[molpro-user] Unclear failure of two molpro tests, ch2_rs2_f12.test and lif_nacme_bo.test

Evgeniy Gromov Evgeniy.Gromov at pci.uni-heidelberg.de
Fri Feb 1 18:38:07 GMT 2013


Dear Andy,

Thank you very much for your efforts in clarifying the problem with
those two tests (ch2_rs2_f12.test and lif_nacme_bo.test). I have finally
recompiled Molpro without -openmp (as you suggested) and using MKL 
instead of ACML. Now all tests have passed OK. So, the problem indeed
had to do with "bugs" (or whatever) in pgi 10.9 and acml (4.4.0).

Again, thanks a lot for your help!

Best wishes,
Evgeniy

Andy May wrote:
> Evgeniy,
>
> For ch2_rs2_f12 I finally managed to reproduce the problem by installing
> pgi 10.9. This problem is not seen with the latest pgi (12.10), so it
> could be a compiler bug that has now been fixed. The problem can be
> avoided with pgi 10.9 by not specifying -openmp. There is not much omp
> code in Molpro, so I think this would be the best option if you don't
> have access to a newer version of pgi. I've opened a bug report about this:
>
> https://www.molpro.net/bugzilla/show_bug.cgi?id=3971
>
> and will see if there is a simple way to avoid this with omp still enabled.
>
> I also opened a bug about lif_nacme_bo failing with ACML:
>
> https://www.molpro.net/bugzilla/show_bug.cgi?id=3967
>
> and I'll attempt to find if this is a Molpro of ACML problem.
>
> Best wishes,
>
> Andy
>
> On 23/01/13 10:07, Evgeniy Gromov wrote:
>> Dear Andy,
>>
>> Many thanks for your response. Enclosed please find the CONFIG file
>> and the ch2_rs2_f12.errout file (output of ch2_rs2_f12.test).
>> Perhaps I will try to re-compile the code with MKL, as it seems to
>> be more stable.
>>
>> Best wishes,
>> Evgeniy
>>
>> Andy May wrote:
>>> Evgeniy,
>>>
>>> For ch2_rs2_f12.test, I wasn't able to reproduce the problem. Can you
>>> send the output file for this job, and the CONFIG file of the build, to
>>> see what the differences between the builds are.
>>>
>>> For lif_nacme_bo.test, I was able to reproduce the failure. It seems to
>>> only occur with ACML, changing to MKL or not using an external
>>> blas/lapack library then the job runs fine. I'm not sure if this is a
>>> bug in ACML, or a bug in Molpro only seen when using ACML, I'll try to
>>> debug.
>>>
>>> Best wishes,
>>>
>>> Andy
>>>
>>> On 20/01/13 16:58, Evgeniy Gromov wrote:
>>>> Dear Developers and Users of Molpro,
>>>>
>>>> I wonder if someone faced problems with the molpro tests
>>>> ch2_rs2_f12.test and lif_nacme_bo.test.
>>>> The former crashes after the 1st MRCI iteration (it is
>>>> a MRCI-F12 calculation), while the latter crashes at the
>>>> step:
>>>> ci;trans,6000.2,6001.2,biorth;
>>>>
>>>> In both cases the error is unclear:
>>>>
>>>> [node022:28675] *** Process received signal ***
>>>> [node022:28675] Signal: Segmentation fault (11)
>>>> [node022:28675] Signal code: Address not mapped (1)
>>>> [node022:28675] Failing at address: 0x201820b528
>>>>
>>>> All the other molpro tests passed OK! So I really wonder
>>>> why those two crashed?
>>>>
>>>> Best regards,
>>>> Evgeniy
>>>>
>>>> ps: Details of the compilation:
>>>>
>>>> compiler=pgf90 (10.9), openmpi (1.2.8), acml (4.4.0)
>>>>
>>>> CONFIGURE_OPTIONS="-i8" "-openmp" "-pgcc" "-pgf90" "-nocuda" "-mpp"
>>>> "-mppbase" "/cvos/shared/apps/openmpi/pgi/64/1.2.8/include" "-blaspath"
>>>> "/cvos/shared/TC/acml/4.4.0/pgi64_mp_int64/lib"
>>>
>>
>>
>


-- 
_______________________________________
Dr. Evgeniy Gromov
Theoretische Chemie
Physikalisch-Chemisches Institut
Im Neuenheimer Feld 229
D-69120 Heidelberg
Germany

Telefon: +49/(0)6221/545263
Fax: +49/(0)6221/545221
E-mail: evgeniy at pci.uni-heidelberg.de
_______________________________________






More information about the Molpro-user mailing list