[Wien] some thoughts after spending a working day with WIEN2K

[Laurence Marks]

Let me add a couple of comments about autoconf/automake, as someone
who has written code which uses them. These scripts are designed to
work around the reality that c/c++ as well as linux systems are rarely
standard, so the developer can design around this as much as possible,
for instance replacing broken malloc or similar subroutines. Support
for fortran in autoconf/automake is minimal, see
http://www.gnu.org/software/autoconf-archive/The-Macros.html#The-Macros
. The only use these scripts would have for Wien2k would be for the c
timers (which is already covered as well as they would do). They would
NOT find the correct libraries much better than is currently done,
particularly as the best Intel compile options change every six months
or so as Intel changes ifort.

If there was much tighter integration between fortran and these
macros, then it would be useful to switch -- they are certainly very,
very useful with c/c++ and some other codes. Until then (which I
expect to be never) the user HAS to determine what is the correct
compiler to use, as well as the correct mpi (which is actually a far
worse issue).

On Sun, Jun 6, 2010 at 2:16 PM, Peter Blaha
<pblaha at theochem.tuwien.ac.at> wrote:
>> Some suggestions after my first (rather frustating) day with wien2k:
>
> As far as I understood from your personal email directly to me (which I
> answered, although the rules clearly say you should post questions in the
> mailing list) the most frustrating problem were:
> i) you had some confusion about default options in your gfortran
> compiler........
> ii) you did not like to follow our suggestions and use ifort/mkl.
>
>> 1) a) Distribute binaries for common architechtures (see
>> http://www.turbomole.com/)
>
> We distribute the binaries. Just download and run them.
>
>>    b) additionally profide a configure script to figure out your
>> system setup and making a basic Makefile.
>>
>>  http://www.gnu.org/software/autoconf/manual/html_node/index.html#Top
>
> We have chosen a different scheme (mainly for historical reasons, there was
> no
> autoconf when this was started and WIEN2k is a very "unconventional" and
> unique
> program).
> Furthermore, I doubt that autoconfigure helps when people do not have
> compiler/libraries
> in their path, mix 32 and 64 bit, .....
> My experience with public domain software products is that one needs to read
> the doku and experiment with options, EVEN IF they use autoconfig.
> Of course, for you personally it would have been easier, since you seem to
> be an
> expert in "free" software products (somebody else should do the work and I
> benefit...)
>
>>    c) check that the installation instructions are upto-date (now in
>> the gfortran installation there is some confusing 'goto' library,
>>        a non-existing www-address for this library is given -> even
>> more confusing)
>
> Maybe the goto webaddress is now outdated, but a simple "goto library" in
> google points directly to the proper libs.
>
> I would expect that using gfortran + Ubuntu-blas  on a modern PC performs
> worse
> than  ifort+mkl on a 5 year old PC!
> If I'm wrong, I'm looking for your contribution of a good benchmark time ...
>
> And the "tip" with gotoblas is probably a very valuable hint .... (even if
> the
> link has changed)
>
>> 2) Provide a proper test cases in order to see whether programs work
>> correctly (see  http://www.turbomole.com/)
>> 3) Provide example inputs (see  http://www.turbomole.com/)
>
> They exists. Of course one can improve everything, and for a commercial
> software thiis is really necessary. But I consider WIEN2k as a product
> for a VERY modest fee and I cannot employ somebody permanently. Or would
> you like to do a "postdoc" with me and as outcome of your postdoc time you
> can say you have written a "perfect documentation for WIEN2k" ??? Maybe you
> can
> tell us how much   turbomole   consts AND in addition one gets
> free program updates twice a year for 10 years now !!
>
>
>> 4) Provide tutorials (see http://ambermd.org/tutorials/)
>> 5) Provide clear documentation (see https://wiki.fysik.dtu.dk/ase/)
>>    This is probably rather ok as it is, but for instance I was not able to
>> use
>>     converter xyz2struct based on the manual, one had to look at the
>> source code.
>
> I try my best. If you can make a specific suggestion or even better,
> write some lines which gives a clearer documentation, I'll be
> happy to include them in the next release.
>
>> 6) It seems that w2web is too easy to use as black box. There should
>> be more links to documentation
>>    directly from the w2web.
>> 7) Mailing list seems to be pretty passive, lots of questions with no
>> answers.
>
> The Mailing list is VERY active, but we have a clear regulation:
> i) Questions which have been answered before (maybe more than 20 times) will
> not be answered again.
> ii) You need to provide enough and proper information, otherwise nobody will
> help. We cannot
>    afford the time to ask for this info or play "Inspector Columbo" to find
> the hidden details.
>
> We are all scientists (chemists and physisists) and mostly interested in a
> scientific problem.
> Very few of us are "computer freaks" (making the "perfect" software") or
> "good angels", who "want"
> to help if somebody is too lazy to read the doku or the previous mails.
>
> A typical example is attached below, which was raised just after your
> complains.
> I'm NOT going to answer this, because this is very detailed explained in the
> UG.
>
>> Probably there is no alternative for this program for certain
>> problems, so paying $$$ may be justified.
>
> For every program there is an alternative. But sometimes it is at the end of
> the
> day cheaper to pay something in the morning....
>
> ---------------------------------------------------------------------------------
> Instead of all those general complains, it would have
> been of great value if you would have provided clear answers to specific
> problems:
>
> How did you finally settle the compilation problems...
> Provide compile options for  gfortran + "your blas"
> Provide benchmark timings.
> Tell exactly what was unclear in the (I agree, very short) docu of
> xyz2struct and
> even better, provide a few lines with a better doku.
> Give the correct web-site of the goto library
> ....
>
> Best regards
> --
> -----------------------------------------
> Peter Blaha
> Inst. Materials Chemistry, TU Vienna
> Getreidemarkt 9, A-1060 Vienna, Austria
> Tel: +43-1-5880115671
> Fax: +43-1-5880115698
> email: pblaha at theochem.tuwien.ac.at
> -----------------------------------------
>
>
> I am sorry I am asking the same question again. In the machines file
>
> If I write
>
> 1:nx80
>
> 1:nx80
>
> 1:nx80
>
> The SCF runs with lapw1para and so on a single node (having 8 processors) in
> 3 cpu’s (cores)
>
>
>
> Now if  write
>
> 1:nx80:3
>
> Granularity:3
>
> Extrafine:1
>
> It crashes
>
>
>
> Can anybody tell me as to how we can make the second one work.
>
> We have written a script but it works on several nodes but one cpu on a
> single node. We are not able to do k-point parallel on several nodes with
> all processors on each node.
>
>
> ANSWER: Please read the UG
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>



-- 
Laurence Marks
Department of Materials Science and Engineering
MSE Rm 2036 Cook Hall
2220 N Campus Drive
Northwestern University
Evanston, IL 60208, USA
Tel: (847) 491-3996 Fax: (847) 491-7820
email: L-marks at northwestern dot edu
Web: www.numis.northwestern.edu
Chair, Commission on Electron Crystallography of IUCR
www.numis.northwestern.edu/
Electron crystallography is the branch of science that uses electron
scattering and imaging to study the structure of matter.