The ABINIT Users Mailing List ( CLOSED )

[gilles zerah <gilles.zerah@wanadoo.fr>]

Dear Andreas,
I looked into your case.
It yields ngfft 40 40 320 and 63 bands. You might try another band/fft 
parallelization.
This approach relies on MPI_ALLTOALL, so you need a good network
and a good implementation of MPI.

To try this different approach of  the band parallelisation:

-compile a recent version (e.g. 5.3.4) with MPI_FFT flag instead MPI

-perform a permutation on your data (acell, kptrlatt, xred etc..) to 
bring z in place of x

-use the flags indicated under the keyword npband.
You might want to try different combinations of npband and npfft.

Tell us what happens,

Have fun,

Gilles
> Dear abiniters,
>
> I hope you can help me with some issues with the band-parallelisation 
> using the variables "wfoptalg" and "nbdblock".
> Calculations without this band-parallelisation run normally with nkpt 
> = 6, but take really long. To speed my calculations up, I chose to use 
> the band parallelisation option of abinit,
> wfoptalg 1
> nbdblock 2,
> which makes it, in my case, possible to use 12 processors.
>
> Sadly this parallelisations doesn´t work properly.
> The error-message I get is
> "forrtl: severe (174): SIGSEGV, segmentation fault occurred"
>
>
> So I tested the differences in the three following cases:
>
> So, without parallelisation
> # wfoptalg 1
> # nbdblock 2
> -> calculation is running
>
> With parallelisation-variables, but calculating one band per block 
> (equals no parallelisation)
> wfoptalg 1
> nbdblock 1
> -> calculation is running
>
> With parallelisation, two bands per block
> wfoptalg 1
> nbdblock 2
> -> error-message "forrtl: severe (174): SIGSEGV, segmentation fault 
> occurred"
> (This message appears for every processor)
>
>
> I hope you can help me to get this working,
>
> best regards
> Andreas Boehner
>
>
> PS:
> INPUT-FILE:
> chkprim 0
> tsmear 0.01
> occopt 7
> fband 0.5
> prtden 1
> #irdwfk 1
>
> *wfoptalg 1
> nbdblock 2*
>
> # Positions-Optimierung
> # ionmov 3
> # ecutsm 0.5
> # ntime 10
>
>
> # EZ
> acell 7.48 7.48 58.588468
>
> rprim
>  1   0   0
>  0   1   0
>  0   0   1
>
> # Atomsorte
> ntypat 1
> znucl 24
>
> # Aufbau des Gitters
> natom 18
> typat 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1  xred
>  0   0   0           # Ebene 1  Flaeche
>  0.5 0.5 0
>  0.5 0   0.0451627   # Ebene 2  Innenzentriete Atome
>  0   0.5 0.0451627
>  0   0   0.0903255   # Ebene 3
>  0.5 0.5 0.0903255
>  0.5 0   0.1354882   # Ebene 4
>  0   0.5 0.1354882
>  0   0   0.180651    # Ebene 5
>  0.5 0.5 0.180651
>  0.5 0   0.2258138   # Ebene 6  0   0.5 0.2258138
>  0   0   0.2709765   # Ebene 7
>  0.5 0.5 0.2709765
>  0.5 0   0.3161393   # Ebene 8
>  0   0.5 0.3161393
>  0   0   0.3613021   # Ebene 9  Cr-bcc Ende
>  0.5 0.5 0.3613021
>
> # Konvergenzparamter
> ecut 35
> kptopt 1
> #prtkpt 1
> kptrlatt    # 6 kpt  kptrlen = 44.88xxx
> 6 0 0 0 6 0 0 0 1
>
> nshiftk 1
> shiftk 0.5 0.5 0.5
>
> # SCF
> nstep 30
> toldfe 1.0d-6