[Wien] wien2k, gotoblas and multi threads

[Todd Pfaff]

Yang,

According to answers I've gotten from Dr. Peter Blaha on the wien2k
users mailing list, see below for what I believe we are supposed to be
doing in order to run a wien2k process in parallel on a single machine
using multiple cpu cores.  I suggest that if you have further questions
about any of this that you subscribe to that mailing list and ask the
questions there yourself.

   - create a .machines file in the working directory with N of these
   lines where N is the number of processors you want to use:

     1:localhost

   - I believe you may also want a "residue" line as described in the
   User's Guide.  I don't know if any of the other .machines file
   directives are necessary or desireable in this case.

   - run 'x lapw1 -p'.

Apparently this k-point parallelism .machines file setup is supposed to
be obvious from the wien2k user's guide, but I've read the relevant parts
of the user's guide and the FAQ several times and I can't find anywhere
that this is described clearly.

Dr. Blaha has also advised me that we will get much better performance
from wien2k if we were to use the Intel ifort compiler and Intel MKL
library, instead of gfortran and GotoBLAS.  My test results agree.  I'm
testing this now using the free downloads from Intel, but if we want to
continue using the Intel tools them we'll have to purchase them.  We
probably shouldn't have even bothered wasting time with gfortran and
GotoBLAS given the huge difference I'm seeing in performance, but they're
mentioned in the wien2k users guide so I assumed they would be suitable
alternatives.  It's possible that we could squeeze a bit more performance
from gfortran+gotoblas with some more optimization but I'm doubtful that
the extra investment in time will pay off.


--
Todd Pfaff <pfaff at mcmaster.ca>
Research & High-Performance Computing Support
McMaster University, Hamilton, Ontario, Canada
http://www.rhpcs.mcmaster.ca/~pfaff