[CP2K:2723] compiling on XT5

[ivaylo]

Hi all,
Has anyone encountered the following when trying to compile on Cray
XT5?

------------------
ftn -c -freeform -I/opt/fftw/3.1.1/cnos/include/
qs_integrate_potential.f90
/opt/cray/xt-asyncpe/3.7/bin/ftn: INFO: linux target is being used

MODULE qs_integrate_potential
^
pathf95-855 pathf90-3.2.99: ERROR QS_INTEGRATE_POTENTIAL, File =
qs_integrate_potential.f90, Line = 30, Column = 8
The compiler has detected errors in module
"QS_INTEGRATE_POTENTIAL". No module information file will be created
for this module.

CALL dbcsr_distribution_make_threads(dist)
^
pathf95-1688 pathf90-3.2.99: ERROR INTEGRATE_V_RSPACE, File =
qs_integrate_potential.f90, Line = 869, Column = 42
Because dummy argument DIST is volatile, this actual argument is not
allowed

CALL dbcsr_distribution_make_threads(dist)
----------------
Thanks,
Ivaylo


Compiling with pathscale on XT5 and the following settings:
-------------------
CC = cc
CPP = cpp
FC = ftn
LD = ftn
AR = ar -r
DFLAGS = -D__XT5 -D__FFTW3 -D__FFTSG -D__parallel -D__BLACS -
D__SCALAPACK -D__NO_STATM_ACCESS
CPPFLAGS = -traditional -C $(DFLAGS) -P -I/opt/fftw/3.1.1/cnos/include
FCFLAGS = -freeform -I/opt/fftw/3.1.1/cnos/include
LDFLAGS = $(FCFLAGS) -static
LIBS = -L/opt/fftw/3.1.1/cnos/lib -lfftw3 -lfftw3f
-------------------

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To post to this group, send email to cp...@googlegroups.com.
To unsubscribe from this group, send email to cp2k+uns...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/cp2k?hl=en.

[IBethune]

This is now fixed in CVS. Thanks Urban!

[ivaylo]

Hi,
Many thanks for fixing the problem with compiling cp2k on the XT5.
Can I also ask for some advice on the best ways to achieve scaling
on the XT5 platform? Any pointers/suggestions would be greatly
appreciated.

Regards,
Ivaylo

> For more options, visit this group athttp://groups.google.com/group/cp2k?hl=en.- Hide quoted text -
>
> - Show quoted text -

[Axel]

On May 24, 11:56 pm, ivaylo <ivo6iva...@gmail.com> wrote:
> Hi,
> Many thanks for fixing the problem with compiling cp2k on the XT5.
> Can I also ask for some advice on the best ways to achieve scaling
> on the XT5 platform? Any pointers/suggestions would be greatly
> appreciated.

you best shot is probably using one MPI task per processor socket
(i.e. two MPI tasks per node) and then experiment with OpenMP.

that is what i found with classical MD codes and was confirmed
by people using finite element methods.

cheers,
axel.

> > For more options, visit this group athttp://groups.google.com/group/cp2k?hl=en.-Hide quoted text -

>
> > - Show quoted text -
>
> --
> You received this message because you are subscribed to the Google Groups "cp2k" group.
> To post to this group, send email to cp...@googlegroups.com.
> To unsubscribe from this group, send email to cp2k+uns...@googlegroups.com.

[IBethune]

There is always going to be a trade-off between best performance and
efficient use of resources (i.e. cost/CPU-hours).

Generally, Pathscale is found to give the best performance of the
compilers on the XT, and you should use the most recent xt-libsci
(BLAS, Sca/LAPACK etc.), fftw and xt-mpt modules, as these do provide
real performance benefits over older versions.

Using only a sub-set of the cores on the node may give some
improvement, and if you want to use OpenMP, make sure you are using a
recent version from CVS as this is under constant development.

- Iain

> > For more options, visit this group athttp://groups.google.com/group/cp2k?hl=en.-Hide quoted text -