[Pw_forum] computation time of addusdens subroutine

[Paolo Giannozzi]

The postprocessing code was written having in mind the case in which one is
interested to a single wavefunction or a small group of wavefunctions. If
you use it for all wavefunctions, it will take a lot of time, since it
calls 'addusdens' many times instead of one. The only solution is to modify
the postprocessing code according to your needs.

Paolo

On Tue, Jul 11, 2017 at 4:04 PM, Junfeng Qiao <qiao0junfeng at gmail.com>
wrote:

> Dear QE users,
>
> I am currently trying to extract wavefunction |psi|^2 of some bands and
> kpoints, which can be achieved with plot_num = 7 of pp.x.
>
> However, if I extract large amount of wavefunctions, for example, assigning
> kpoint(1) = 1, kpoint(2) = 20*20, kband(1) = 1, kband(2) = 300,
> the pp.x will run for an unbearable long time.
>
> On my cluster (10 intel cpus, total 60 cores), each wavefunction will take
> about 8 seconds, and do a simple prediction, in total 8 * (20*20) * 300 /
> 3600
> / 24 = 11 days ! It is much more longer than the scf calculations.
>
> After playing with the code(QE v6.1), I found in local_dos.f90 line 392,
> the
> call to addusdens() takes up almost all the computation time, other parts
> are
> negligible.
>
> While, in this situation, I totally have no ideas how to get out of this
> trouble, so I am seeking for help in this community: is it possible to
> speed
> up this addusdens() part?
>
> Thanks for your time of reading this email, any suggestions will be truly
> appreciated.
>
> Sincerely,
> Junfeng Qiao
> Beihang University (BUAA)
> Beijing, China
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum
>



-- 
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/users/attachments/20170712/9bc9f6ff/attachment.html>