[QE-developers] Performance Optimization of QE

[Carlo Cavazzoni]

Dear Tobias,

I had a quick look to the poster you presented @ SC18,
and actually the work you did on the FFT is quite close to what
we are doing to improve/remove the Task Group approach
that was developed many years ago in collaboration with IBM Zurich both 
for CP and CPMD...
In particular there is new contribution coming from a collaboration with 
Intel
that should integrate the work we did for GPU, and containing the overlap
between batches of FFT and communication, using OpenMP tasks.
So I suggest to find some time slot to discuss how to better integrate 
the different/similar
approaches to the FFT into CP (and in general QE, since the FFT driver 
is one for all the QE module).
Much interesting also the work on the linear algebra, for which we are 
not working yet on an
overlap strategy, but rather we worked for a strategy of two nested 
parallelization (in the orthogonalization),
which reduce the amount of data to be exchanged (e.g. if you have only 
the parallelization over G vector)

best,
carlo

Il 27/11/2019 13:02, Tobias Kloeffel ha scritto:
> Dear Paolo,
>
> yes I do know the differences and I am also aware of some differences 
> in the implementation of QE CP / CPMD
>
> Either just to the CP code of QE or in general to QE.
>
> However, QE is more 'alive' than CPMD so if you are interested we 
> should try to get to some sort of workplan, discuss what changes you 
> would accept and what not.
>
>
>
>
> On 11/27/19 12:51 PM, Paolo Giannozzi wrote:
>> Interesting, but ... it seems to me that you have been working on 
>> "CPMD" from www.cpmd.org <http://www.cpmd.org>. This is a different 
>> project, not part of Quantum ESPRESSO. There is a similar 
>> Car-Parrinello code called "CP" in QE. Are you considering porting 
>> your improvements to the CP code of QE as well? that would be great
>>
>> Paolo
>>
>>
>> On Wed, Nov 27, 2019 at 12:44 PM Tobias Kloeffel 
>> <tobias.kloeffel at fau.de <mailto:tobias.kloeffel at fau.de>> wrote:
>>
>>     Dear QE Developers,
>>     in the last two years I reworked the USPP code of CPMD and obtained
>>     quite impressive results. E.g.
>>     https://sc18.supercomputing.org/proceedings/tech_poster/tech_poster_pages/post189.html
>>
>>     Please also find some updated performance plots on SUPERMUC-NG
>>     (Skylake
>>     48 cores/Node):
>>
>>     NCPP/USPP comparison for the current CPMD 4.3 and my reworked version
>>     timestep  = 0.145 fs, 256 water molecules, 1024 states,
>>     USPP: 2560 beta projectors, 25 Ry
>>     NCPP: 512 beta projectors, 80 Ry
>>     4.3: PS_per_day_nc-old-ng.pdf
>>     rev: PS_per_day_nc-new-ng.pdf
>>
>>     and some more water benchmarks from 32-2048 molecules.
>>
>>
>>     Currently I am thinking of applying the same optimizations to QE,
>>     so the
>>     big question is what are your thoughts about this?
>>     I had to rework quite a large portion of the code (~ 14625
>>     additions and
>>     6576 deletions) so the question is really how I could proceed:
>>     a) Fork QE and just rewrite the parts of the code that are
>>     important on
>>     my side
>>     b) Fork QE and try to keep everything mergeable
>>     c) Directly create small patches
>>
>>     So for now, please just let me know if there is some interest on
>>     your side.
>>
>>     Kind reagrds,
>>     Tobias Klöffel
>>
>>
>>     -- 
>>     M.Sc. Tobias Klöffel
>>     =======================================================
>>     Interdisciplinary Center for Molecular Materials (ICMM)
>>     and Computer-Chemistry-Center (CCC)
>>     Department Chemie und Pharmazie
>>     Friedrich-Alexander-Universität Erlangen-Nürnberg
>>     Nägelsbachstr. 25
>>     D-91052 Erlangen, Germany
>>
>>     Room: 2.305
>>     Phone: +49 (0) 9131 / 85 - 20423
>>     Fax: +49 (0) 9131 / 85 - 26565
>>
>>     =======================================================
>>
>>     E-mail: tobias.kloeffel at fau.de <mailto:tobias.kloeffel at fau.de>
>>
>>     _______________________________________________
>>     developers mailing list
>>     developers at lists.quantum-espresso.org
>>     <mailto:developers at lists.quantum-espresso.org>
>>     https://lists.quantum-espresso.org/mailman/listinfo/developers
>>
>>
>>
>> -- 
>> Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
>> Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
>> Phone +39-0432-558216, fax +39-0432-558222
>>
>
> -- 
> M.Sc. Tobias Klöffel
> =======================================================
> Interdisciplinary Center for Molecular Materials (ICMM)
> and Computer-Chemistry-Center (CCC)
> Department Chemie und Pharmazie
> Friedrich-Alexander-Universität Erlangen-Nürnberg
> Nägelsbachstr. 25
> D-91052 Erlangen, Germany
>
> Room: 2.305
> Phone: +49 (0) 9131 / 85 - 20423
> Fax: +49 (0) 9131 / 85 - 26565
>
> =======================================================
>
> E-mail:tobias.kloeffel at fau.de
>
> _______________________________________________
> developers mailing list
> developers at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/developers


-- 
Ph.D. Carlo Cavazzoni
Business Unit High Performance Computing
CINECA - Via Magnanelli 6/3, 40033 Casalecchio di Reno (Bologna)
Tel: +39 051 6171411  Fax: +39 051 6132198
www.cineca.it

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20191127/fe6f57d0/attachment-0001.html>