[Q-e-developers] global_kpoint_index

Lorenzo Paulatto paulatz at gmail.com
Wed Oct 19 22:03:24 CEST 2016


Hi Paolo, thank you for answering. I've actually solved this yesterday,
introducing a set of communications orthogonal to the pools, which turned
out to be relatively easy. However, since I went on holiday this morning,
and because the code needs some more testing, I won't uploas it before next
week! If anybody wants to try it, I may be able to put it somewhere..

Cheers

-- 
Lorenzo Paulatto
Written on a virtual keyboard with real fingers

On 19 Oct 2016 2:48 p.m., "Paolo Giannozzi" <paolo.giannozzi at uniud.it>
wrote:

> On Tue, Oct 18, 2016 at 11:52 AM, Lorenzo Paulatto <
> lorenzo.paulatto at impmc.upmc.fr> wrote:
>
> Yes, I see, I had to redo all the scatter/collect mechanism anyway because
>> <beta|psi> products are stored in a derived type, which has not pre-made
>> subroutines.
>>
>
> likely you will need a complex version. maybe we should consider an
> interface.
>
> I also noticed that we do:
>>     IF (npool>1) CALL mp_sum(exxbuff, inter_pool_comm)
>> should this be followed by a
>>  CALL mp_bcast(exxbuff, root_pool, intra_pool_comm)
>> ?
>>
>
> I don;'t think so: exxbuff has three indices and is distributed over the
> first (PW, inside a band group) and the second (bands, across band groups)
> indices. The problem is the third index, k+q. Each pool has a subset of
> k-points, fills a subset of k+q points, needs a different subset of k+q
> points (except if nq1=nq2=nq3=1, that is, q=0 only). The "mp_sum" above is
> just filling the third index for all pools. It is dumb but simple; better
> solutions welcome
>
> Paolo
> --
> 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
>
>
> _______________________________________________
> Q-e-developers mailing list
> Q-e-developers at qe-forge.org
> http://qe-forge.org/mailman/listinfo/q-e-developers
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20161019/208eaee9/attachment.html>


More information about the developers mailing list