Weird forces

Kyle kyle... at gmail.com
Fri Mar 19 23:12:00 UTC 2010


Hi Toon,

Thanks for your reply.

We've also tried the serial version and it worked fine for us.  So
only the MPI version is giving problems.  We'll keep trying different
libraries to find out the problem.

Best regards,
Kyle


On Mar 17, 3:10 pm, Toon Verstraelen <Toon.Ver... at UGent.be>
wrote:
> Hi Kyle,
>
> I'd suggest to start from the most basic compilation one can think of:
> serial, blas and lapack from netlib.org, gfortran or g95, fftsg.
> Increase the complexity of your compilation step by step and check which
> step causes the failure by repeating the regtests. The script
> do_regtests in the tools directory is worth reading. That will at least
> point out the guilty. Then we can start thinking of a solution.
>
> This is an arch file I give to students to compile a serial cp2k on a
> regular ubuntu system. It should be pretty generic.
>
> CC       = cc
> CPP      =
>
> FC       = gfortran
> LD       = gfortran
>
> AR       = ar -r
>
> CPPFLAGS =
> DFLAGS   = -D__GFORTRAN -D__FFTSG
> FCFLAGS  = -O2 -ffast-math -funroll-loops -ftree-vectorize -march=native
> -ffree-form -ffree-line-length-none $(DFLAGS)
> LDFLAGS  = $(FCFLAGS)
> LIBS     = -lblas-3 -llapack-3
>
> OBJECTS_ARCHITECTURE = machine_gfortran.o
>
> This won't break speed records, but it doesn't have all the fancy
> glitter that tends to make compilations more difficult and unreliable.
> This should give the same result as Teo had.
>
> cheers,
>
> Toon
>
> Kyle wrote:
> > Dear experts,
>
> > Just following up the previous post.  It might be hard to find out the
> > problem from the architecture information, but if possible, could you
> > provide examples that worked well so that we can try to change the
> > libraries and do tests?
>
> > Thanks,
>
> > Best regards,
> > Kyle



More information about the CP2K-user mailing list