[Debian-science-sagemath] Problems with CPU extensions used by dependencies
Tobias Hansen
thansen at debian.org
Tue Aug 9 19:53:06 UTC 2016
Ok, you can also just run
grep --color -P "avx2|fma|mmx|sse2|sse4_1|sse" /proc/cpuinfo
and see if you find these 6 extensions in the list.
I was able to build Sage 7.3 without speaklater and the ptestlong.log
also does not look like speaklater is missing for anything.
Best,
Tobias
On 08/09/2016 08:26 PM, Jerome BENOIT wrote:
>
>
> On 09/08/16 19:55, Tobias Hansen wrote:
>> I checked now and I have all of these extensions available. That could
>> explain why less tests are failing for me. Jerome, did you already run
>> that one-liner to check?
>
> Because of my zombie story (yes, I believe in zombies), I postpone the test.
>
> Jerome
>
> PS:
> 1] I closed the ITP for speaklater
> 2] but I maintain my request to be member of the Python list.
>
>
>> Best,
>> Tobias
>
>
>> On 08/09/2016 02:46 AM, Ximin Luo wrote:
>>> It might be that Sage is adding extra optimisations not supported by our CPUs. For example, running this from sagemath.git after a failed test:
>>>
>>> $ x=$(grep -o ' -m[^ ]*' sage/logs/pkgs/sagelib-7.2.log | sort -u | tee /dev/fd/2 | cut -b4- | tr '\n' '|'); grep --color -P "${x%|}" /proc/cpuinfo
>>> -mavx2
>>> -mfma
>>> -mmmx
>>> -msse
>>> -msse2
>>> -msse4.1
>>> flags : [..] mmx fxsr sse sse2 [..] ssse3 [..] sse4_1 sse4_2 [..]
>>> [..]
>>>
>>> I can see that avx2 fma are missing for me at least.
>>>
>>> The difficulty for Debian is that we need to support all i386/amd64/etc CPUs, so we basically have to switch off all optimisations like this.
>>>
>>> X
>>>
>>> Jerome BENOIT:
>>>> Hi,
>>>>
>>>> On 08/08/16 19:45, Tobias Hansen wrote:
>>>>> I don't know. I think nobody packaged it yet because it's a relatively
>>>>> new dependency of Sage. And that things fail is because of the hacky way
>>>>> we are using it currently.
>>>>
>>>> I will think about it. Meanwhile I submitted an ITP for SageTeX.
>>>>
>>>> Thanks,
>>>> Jerome
>>>>
>>>>
>>>>> Best,
>>>>> Tobias
>>>>
>>>>> On 08/08/2016 06:52 PM, Jerome BENOIT wrote:
>>>>>> Is cysignals a very tricky/involve Cyphon software package ?
>>>>>>
>>>>>> Jerome
>>>>>>
>>>>>> On 08/08/16 12:59, Tobias Hansen wrote:
>>>>>>> Hi,
>>>>>>
>>>>>>> so there still seem to be big problems with cysignals:
>>>>>>
>>>>>>> 877x
>>>>>>> File "src/cysignals/signals.pyx", line 100, in
>>>>>>> cysignals.signals.sig_raise_exception (build/src/cysignals/signals.c:1225)
>>>>>>> SignalError: Illegal instruction
>>>>>>
>>>>>>> Best,
>>>>>>> Tobias
>>>>>>
>>>>>>> On 08/07/2016 02:27 PM, Jerome BENOIT wrote:
>>>>>>>> Please find in attachment my last log files:
>>>>>>>> the number of Failed examples is 1949
>>>>>>>>
>>>>>>>> Jerome
>>>>>>>>
>>>>>>>> On 05/08/16 11:57, Tobias Hansen wrote:
>>>>>>>>> On 08/05/2016 11:46 AM, Jerome BENOIT wrote:
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On 05/08/16 11:25, Tobias Hansen wrote:
>>>>>>>>>>> Hi,
>>>>>>>>>>
>>>>>>>>>>> the first two errors I noticed were
>>>>>>>>>>
>>>>>>>>>>> 1839x
>>>>>>>>>>> In order to initialize the database,
>>>>>>>>>>> /home/jgmb/.local/share/sagemath/conway_polynomials/conway_polynomials.sobj
>>>>>>>>>>> must exist.
>>>>>>>>>>
>>>>>>>>>>> 999x
>>>>>>>>>>> Desired database (='cremona mini') does not exist
>>>>>>>>>>
>>>>>>>>>>> So it looks in /home/jgmb/.local/share/sagemath instead of
>>>>>>>>>>> /usr/share/sagemath, where for example
>>>>>>>>>>> conway_polynomials/conway_polynomials.sobj is installed by the package
>>>>>>>>>>> sagemath-database-conway-polynomials. I think this is due to
>>>>>>>>>>
>>>>>>>>>>> -export SAGE_SHARE="/usr/share/sagemath"
>>>>>>>>>>> +export SAGE_SHARE="$HOME/.local/share/sagemath"
>>>>>>>>>>
>>>>>>>>>>> in debian-temp-userdir.patch.
>>>>>>>>>>
>>>>>>>>>>> I don't understand why Ximin and I don't get these errors, but it should
>>>>>>>>>>> be fixed by cleaning up the mixed /usr and $HOME situation.
>>>>>>>>>>
>>>>>>>>>>> BTW, my failed tests count went up from ~200 to 423 because now some
>>>>>>>>>>> files from cysignals are not found.
>>>>>>>>>>
>>>>>>>>>> please let me know when the git material is updated in view to make new test
>>>>>>>>>> (and getting familiar with the process).
>>>>>>>>
>>>>>>>>> I will probably not have time to work on that today. It's complicated
>>>>>>>>> because it's hard to know for what exactly all the variables such as
>>>>>>>>> SAGE_LOCAL etc are used for. If you want you can also try it to get
>>>>>>>>> familiar with the stuff. In that case I could write a bit more about
>>>>>>>>> what to do.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>> BTW2, SageMath 7.3 was released. Obviously with Singular 3.1.7. But at
>>>>>>>>>>> least when they get that Singular 4 ticket done, we can backport the
>>>>>>>>>>> patch. Or maybe SageMath 7.4 comes soon enough to target that for
>>>>>>>>>>> Stretch. I guess we should fix your huge number of failing doctests
>>>>>>>>>>> before upgrading to SageMath 7.3
>>>>>>>>>>
>>>>>>>>>> Meanwhile we may have a working libgap 4.8.3 packages.
>>>>>>>>>>
>>>>>>>>>> Jerome
>>>>>>>>
>>>>>>>>> Very good. After uploading the git repo you can also try pushing to the
>>>>>>>>> apt and update the sagemath git to use it. The configure.ac in
>>>>>>>>> debian/pruner needs to be updated in order to find it, so you can get
>>>>>>>>> familiar with that.
>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>> Best,
>>>>>>>>>>> Tobias
>>>>>>>>>>
>>>>>>>>>>> On 08/05/2016 10:54 AM, Jerome BENOIT wrote:
>>>>>>>>>>>> Hello Tobias,
>>>>>>>>>>>>
>>>>>>>>>>>> please find in attachment my last log files.
>>>>>>>>>>>>
>>>>>>>>>>>> Jerome
>>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>
>>>>
>>>>
>>>
>
>
>> _______________________________________________
>> Debian-science-sagemath mailing list
>> Debian-science-sagemath at lists.alioth.debian.org
>> https://lists.alioth.debian.org/mailman/listinfo/debian-science-sagemath
>
>
>
> _______________________________________________
> Debian-science-sagemath mailing list
> Debian-science-sagemath at lists.alioth.debian.org
> https://lists.alioth.debian.org/mailman/listinfo/debian-science-sagemath
>
More information about the Debian-science-sagemath
mailing list