[Wien] FFTW3

[Peter Blaha]

FFTW is not really time-critical in WIEN2k.

If you have only a single 16 node machine, mpi for lapw0 us not 
necessary (it might be for lapw1/2; but this depends on the systems you 
want to calculate).
What helps is to compile lapw0 with openmp parallelization, so an 
fftw-omp version is beneficial.
lapw0 parallelizes pretty well with omp, also on 16 cores (provided you 
have at least a medium sized case).
(eventually you can put   omp_lapw0:16 into .machines when you are using 
this computer alone with a single job).

You also need a fftw version which is "compatible" with your fortran 
compiler. If you are using ifort or gfortran, you need to use 
fftw-versions compiled with the corresponding compilers.

PS: Are you sure you have 16 "physical" cores, not just 16 virtual cores 
and only 8 real ones ?

Am 25.02.2022 um 02:21 schrieb Arthur H. Edwards:
> I'm installing W2k on a machine with several fftw libraries. Some are for double precision, some are threaded, some have an omp at the end of ther names, some are static and some are shared. So, under FFTW Settings, when I'm choosing a value for N (FFTW_LIBNAME), it seems a little confusing. Any guidance would be appreciated. As there are only 16 cores on the machine, and it is not part of a larger network of compute nodes, I'm fairly certain that I don't need mpi, but I would be happy to be shown to be wrong.
> 
> Art Edwards
> 

-- 
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-165300             FAX: +43-1-58801-165982
Email: blaha at theochem.tuwien.ac.at    WIEN2k: http://www.wien2k.at
WWW:   http://www.imc.tuwien.ac.at
-------------------------------------------------------------------------