Dear Virginie,

to spot where the problem is you might want to start from the bottom instead of 
starting from the top. I think almost all the examples provided in the QE 
distribution can run in serial on a decent workstation. If I were you I will 
try these options first

1) GNU compilers in serial, without external libraries (editing make.sys and 
replacing -O2 or -O3 with -O0)
$env FC=gfortran F90=gfortran CC=gcc ./configure  --disable-parallel 
--disable-openmp --with-internal-blas --with-internal-lapack

2) Intel compilers in serial, without external libraries (editing make.sys and 
replacing -O2 or -O3 with -O0)
env FC=ifort F90= ifort CC=icc ./configure  --disable-parallel --disable-openmp 
--with-internal-blas --with-internal-lapack

3) Intel compilers in serial, with MKL (editing make.sys and replacing -O2 or 
-O3 with -O0)
env FC=ifort F90= ifort CC=icc ./configure  --disable-parallel --disable-openmp

and then try switching in parallel?.

F

On Apr 15, 2013, at 3:50 PM, TRINITE Virginie <virginie.trinite at 
thalesgroup.com> wrote:

> Dear all
> 
> I have investigated a lot of configurations concerning this problem: changing 
> the library fftw3 blas and lapack, changing the version of espresso , 
> compiling on a node and not on the master, changing the optimization level 
> from -O3 to -O2 (I have no -x flags by default)....
> The problem change or disappear for a particular input file, but I have not 
> found a stable situation where I can run all the tests of PW without error 
> with an increasing number of processors (say until 8 processors).
> The more common error is pw frozen, with no output (difficult to debug).
> With the tests of PW, the berry calculation is usually  the first one to fail 
> with message like that:
>       Frome addusden_r :error #   1
>       Expected 44.000000, found 44.44233857: wrong charge, increase ecutrho
> 
> Finally I think that either my machine has a very nasty problem or
> the problem can come from the compiler (I use ifort 12.1.3).
> It is not the first time that I have problem with the compiler, so I wonder 
> which version of ifort can I use without problem?
> Usually I also use the mkl, but I recompile fftw3 and openmpi with ifort. Is 
> it the more stable way to do?
> 
> Thanks in advance for your response
> 
> Virginie Trinit?
> 
> 
> -----Message d'origine-----
> De : pw_forum-bounces at pwscf.org [mailto:pw_forum-bounces at pwscf.org] De 
> la part de Paolo Giannozzi
> Envoy? : mardi 2 avril 2013 21:44
> ? : PWSCF Forum
> Objet : Re: [Pw_forum] PW is frozen
> 
> On Fri, 2013-03-29 at 10:38 +0100, TRINITE Virginie wrote:
> 
>> I find a very strange behavior of pw.x: It run endlessly without 
>> output anything and with no error, and I have  finally to kill it
>> 
>> This behavior appears by using 6 processors for Iron calculation
> 
> your input works for me on 6 processors (more exactly, 6 processes on 2 
> processors), with slightly different pseudopotentials. V.5.0 may run into 
> strange problems on some machines with k-point parallelization (details in 
> the Doc/release-notes file).
> If this is not your case, it is hard to say what the problem can be without 
> the possibility to make a few tests and to see what is really happening and 
> under which exact conditions. Apart from trying a more recent version, you 
> might consider a poor-man (or poor-woman) debugging by inserting print 
> statements in the main program until you figure out where it hangs
> 
> Paolo
> --
> Paolo Giannozzi, Dept. Chemistry&Physics&Environment, Univ. Udine, via delle 
> Scienze 208, 33100 Udine, Italy Phone +39-0432-558216, fax +39-0432-558222 
> 
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum

--
Mr. Filippo SPIGA, M.Sc.
http://filippospiga.me ~ skype: filippo.spiga

?Nobody will drive us out of Cantor's paradise.? ~ David Hilbert

-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
http://pwscf.org/pipermail/pw_forum/attachments/20130415/a1bc8d8b/attachment.html
 

Reply via email to