[QE-developers] Standard plane-wave parallelization

[Paolo Giannozzi]

I think what you propose is in principles feasible, but I am afraid that it
affects so much code that it is not going to happen anytime soon

Paolo

On Thu, Apr 25, 2019 at 1:05 PM Thomas Brumme <thomas.brumme at uni-leipzig.de>
wrote:

> Dear all,
>
> I understand that the standard parallelization in QE is done via the FFTs
> which also results in a quite good speed up AND memory reduction.
> Of course, the communication between CPUs needs to be fast.
>
> Yet, why is the most important number here the 3rd dimension? Is it
> difficult to implement that the most important dimension is actually the
> largest one? Or would it be "easier" to (internally) rotate the system such
> that the 3rd dimension is the largest and then do the computations and
> rotate back?
>
> I'm asking since I have a large slab system in which z ist not the largest
> dimension and thus I get memory problems on the cluster. Rotating
> the system such that the slab is not in xy but in yz I can reduce the
> memory needed per task... So, I can do this manually but it would be
> easier if this is automatically taken care of ;)
>
> Kind regards
>
> Thomas
>
> --
> Dr. rer. nat. Thomas Brumme
> Wilhelm-Ostwald-Institute for Physical and Theoretical Chemistry
> Leipzig University
> Phillipp-Rosenthal-Strasse 31
> 04103 Leipzig
>
> Tel:  +49 (0)341 97 36456
>
> email: thomas.brumme at uni-leipzig.de
>
> _______________________________________________
> developers mailing list
> 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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20190426/05da64b0/attachment.html>