[gpaw-users] gpaw-setup error

Mon Aug 18 20:58:48 CEST 2014

On 08/18/2014 03:50 PM, Robert Warmbier wrote:
> Ah, thank you. Replacing 'icc.py' with icc/mpicc did the job. :-)
>
> I never checked the scalapack version number of the intel compiler. 
> But the icc+mkl is about a factor of 2 faster than any 
> gcc+whatever(not mkl) I tried on machines with AVX. It's also faster 
> on all other Xeons I use, but not as much. Haven't used a non-Xeon in 
> years though, so can only speak for those cpus.
i see differences no more than 10-20% in favor of icc/mkl vs gcc/openblas.
What matters when in planewave mode is fft - make sure you don't use the 
numpy's fallback in case of gcc build
vs mkl's fft when comparing the performance.
See 
http://listserv.fysik.dtu.dk/pipermail/gpaw-developers/2013-November/004278.html

Best regards,

Marcin
> I usually don't have problems with that combination. But I don't use 
> scalapack, so can't comment on that.
>
> Best Robert
>
>
> ------------------------------------------------------------------------
> *From:* Marcin Dulak [Marcin.Dulak at fysik.dtu.dk]
> *Sent:* 18 August 2014 15:26
> *To:* Robert Warmbier; gpaw-users at listserv.fysik.dtu.dk
> *Cc:* gpaw-developers at listserv.fysik.dtu.dk
> *Subject:* Re: [gpaw-users] gpaw-setup error
>
> On 08/18/2014 02:37 PM, Robert Warmbier wrote:
>> Dear Marcin,
>>
>> thanks for the information.
>> Hm, yes, I remember a discussion about the setups. I was considering 
>> linking the pbe setup to pbesol. Just found that this would be 
>> cheating ;-). I know vasp does it like that.
>>
>> If I may ask: Why is it riskier to use a PBEsol setup generated with 
>> the same parameters used for the PBE setup than using the PBE setup 
>> for PBEsol?
>> In the PSL library  (or PSLibrary...) for quantum espresso they 
>> actually use the same set of parameters to generate UPF files for all 
>> kinds of xc-functionals.
> i would start with PBE setups. The parameters found for PBE setups are 
> generally not transferable to other functionals.
> In the best case the generation of a setup for a different functional 
> using PBE parameters will fail, but it may also generate a bad setup 
> without an error
> (bad means it gives incorrect results when used for some 
> systems/properties).
> Using PBE setups has the advantage that you can cite what you used - 
> if you generate PBEsol yourself there is a risk
> that if something in PBEsol generation goes wrong that make the 
> results look right (there is always a chance for this to happen) it 
> will be hard to reproduce.
> Of course make some tests if your properties are OK to use PBE setups.
>>
>> More directed to the developers mailing list:
>> The problem I had with gpaw-setup came from the weird circumstance, 
>> that the Intel compiler setup does not create a _gpaw.so file (or 
>> deletes it after creating). I couldn't find the part in the code, 
>> where the _gpaw.so is created. I would like to track that down. Does 
>> anyone know, how (where in the code) the _gpaw.so is created? The 
>> whole setup procedure seems to be trimmed towards gcc...
> the problem may be in the icc.py wrapper -  no need to use this.
> Please see an example of using intel here:
> https://trac.fysik.dtu.dk/projects/gpaw/browser/trunk/doc/install/Linux/Niflheim/el6-dl160g6-tm-intel-2013.1.117-openmpi-1.6.3-mkl-2013.1.117-sl-hdf5-1.8.10.py
> I must add that we dropped it in favor of gcc + acml due to stability 
> problems and old scalapack distributed by intel.
>
> Best regards,
>
> Marcin
>
>>
>>
>> Best Robert
>>
>>
>> ------------------------------------------------------------------------
>> *From:* Marcin Dulak [Marcin.Dulak at fysik.dtu.dk]
>> *Sent:* 18 August 2014 13:59
>> *To:* Robert Warmbier; gpaw-users at listserv.fysik.dtu.dk
>> *Subject:* Re: [gpaw-users] gpaw-setup error
>>
>> On 08/18/2014 11:58 AM, Robert Warmbier wrote:
>>> Dear all,
>>>
>>> I want to use the PBEsol functional. However, if I set xc='PBEsol', 
>>> gpaw says
>>>
>>> RuntimeError: Could not find paw.PBEsol-setup for "C".
>>>
>>> So, I tried to generate a setup (any setup):
>>>
>>> gpaw-setup C -f PBEsol
>> normally please do not generate setups.
>> Depending on the properties you investigate it may be OK to cheat
>> GPAW to use PBE setups in PBEsol calculations (just ln -s X.PBE.gz 
>> X.PBEsol.gz).
>> This is OK for total energy differences (and i think we have verified 
>> that also for bandgaps - both GGA and GW)
>> - one can exchange GGA setups and functionals more
>> or less freely, but this is not the case for GLLB and (i guess) maybe 
>> other properties.
>> See the attached plot.
>> The total energies become sometimes positive (people are usually 
>> worried about this) because the reference energy does
>> not correspond between the setup and used functional.
>>
>> If you really want to generate setups
>> see 
>> http://listserv.fysik.dtu.dk/pipermail/gpaw-users/2014-August/002859.html
>>>
>>> I get this error message:
>>>
>>> rbw at rbw-desktop:~/gpawruns> gpaw-setup C
>>> Traceback (most recent call last):
>>>   File "/home/rbw/programs/gpaw/tools/gpaw-setup", line 5, in <module>
>>>     from gpaw.atom.gpaw_setup import main
>>>   File 
>>> "/home/rbw/programs/gpaw/build/lib.linux-x86_64-2.7/gpaw/__init__.py", 
>>> line 239, in <module>
>>>     from gpaw.aseinterface import GPAW
>>>   File 
>>> "/home/rbw/programs/gpaw/build/lib.linux-x86_64-2.7/gpaw/aseinterface.py", 
>>> line 11, in <module>
>>>     from gpaw.xc import XC
>>>   File 
>>> "/home/rbw/programs/gpaw/build/lib.linux-x86_64-2.7/gpaw/xc/__init__.py", 
>>> line 1, in <module>
>>>     from gpaw.xc.libxc import LibXC
>>>   File 
>>> "/home/rbw/programs/gpaw/build/lib.linux-x86_64-2.7/gpaw/xc/libxc.py", 
>>> line 1, in <module>
>>>     import _gpaw
>>> ImportError: No module named _gpaw
>> probably PYTHONPATH not set correctly - it must include the directory 
>> containing _gpaw.so.
>> i guess only gpaw-python works now.
>>>
>>> That is using a current svn 11806 checkout.
>>>
>>> Normal calculations run just fine. Nevertheless, it could be just a 
>>> problem with my configuration:
>>>
>>> I use this customize.py running 'python setup.py build':
>>
>> Best regards,
>>
>> Marcin
>>> compiler = 'icc.py'
>>> libraries  = ['fftw3']   # this should be removed - GPAW does not 
>>> link to it
>>
>>> libraries += ['mkl_intel_lp64', 'mkl_core', 'mkl_sequential', 
>>> 'pthread', 'm']
>>> library_dirs += 
>>> ['/home/rbw/programs/installs/openmpi-1.8.1-intel2013.1.3/lib64']
>>> library_dirs += ['/opt/intel/mkl/lib/intel64']
>>> library_dirs += 
>>> ['/home/rbw/programs/installs/fftw-3.3.4-intel2013.1.3/lib64'] # 
>>> also not necessary
>>
>>> include_dirs += 
>>> ['/home/rbw/programs/installs/openmpi-1.8.1-intel2013.1.3/include']
>>> include_dirs += ['-I/opt/intel/mkl/include']
>>> include_dirs += 
>>> ['/home/rbw/programs/installs/fftw-3.3.4-intel2013.1.3/include']
>>> extra_link_args += ["-fPIC",'-lpthread','-lm']
>>> extra_compile_args += 
>>> ['-I/opt/intel/mkl/include','-static','-std=c99','-fPIC']
>>> mpicompiler = 'icc.py'
>>> mpilinker = 'icc.py'
>>> hdf5 = False
>>> scalapack = False
>>> define_macros += [("GPAW_ASYNC",1)]
>>> define_macros += [("GPAW_MPI2",1)]
>>> include_dirs += 
>>> ['/home/rbw/programs/installs/libxc-2.2.0-intel2013.1.3/include']
>>> library_dirs += 
>>> ['/home/rbw/programs/installs/libxc-2.2.0-intel2013.1.3/lib64']
>>> if 'xc' not in libraries: libraries.append('xc')
>>>
>>> Does anyone have any suggestions for me?
>>>
>>>
>>> Best Robert
>>>
>>>
>>> This communication is intended for the addressee only. It is 
>>> confidential. If you have received this communication in error, 
>>> please notify us immediately and destroy the original message. You 
>>> may not copy or disseminate this communication without the 
>>> permission of the University. Only authorised signatories are 
>>> competent to enter into agreements on behalf of the University and 
>>> recipients are thus advised that the content of this message may not 
>>> be legally binding on the University and may contain the personal 
>>> views and opinions of the author, which are not necessarily the 
>>> views and opinions of The University of the Witwatersrand, 
>>> Johannesburg. All agreements between the University and outsiders 
>>> are subject to South African Law unless the University agrees in 
>>> writing to the contrary.
>>>
>>>
>>>
>>> _______________________________________________
>>> gpaw-users mailing list
>>> gpaw-users at listserv.fysik.dtu.dk
>>> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>>
>>
>> This communication is intended for the addressee only. It is 
>> confidential. If you have received this communication in error, 
>> please notify us immediately and destroy the original message. You 
>> may not copy or disseminate this communication without the permission 
>> of the University. Only authorised signatories are competent to enter 
>> into agreements on behalf of the University and recipients are thus 
>> advised that the content of this message may not be legally binding 
>> on the University and may contain the personal views and opinions of 
>> the author, which are not necessarily the views and opinions of The 
>> University of the Witwatersrand, Johannesburg. All agreements between 
>> the University and outsiders are subject to South African Law unless 
>> the University agrees in writing to the contrary.
>>
>
> This communication is intended for the addressee only. It is 
> confidential. If you have received this communication in error, please 
> notify us immediately and destroy the original message. You may not 
> copy or disseminate this communication without the permission of the 
> University. Only authorised signatories are competent to enter into 
> agreements on behalf of the University and recipients are thus advised 
> that the content of this message may not be legally binding on the 
> University and may contain the personal views and opinions of the 
> author, which are not necessarily the views and opinions of The 
> University of the Witwatersrand, Johannesburg. All agreements between 
> the University and outsiders are subject to South African Law unless 
> the University agrees in writing to the contrary.
>

-- 
***********************************

Marcin Dulak
Technical University of Denmark
Department of Physics
Building 307, Room 229
DK-2800 Kongens Lyngby
Denmark
Tel.: (+45) 4525 3157
Fax.: (+45) 4593 2399
email: Marcin.Dulak at fysik.dtu.dk

***********************************

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.fysik.dtu.dk/pipermail/gpaw-users/attachments/20140818/4388d90d/attachment-0001.html>