new collection of cp2k hacks for intel/infiniband clusters.

lml... at lml... at
Sat Sep 22 22:55:57 UTC 2007


These days, I meet the problem that cp2k just run just one or two
steps, then it hangs there for hours without writing anything, even
errors . Then I patch Axel' hacks as suggested by Axel, now it can
run. The  subroutine m_mov() part really need Alex's patch for the
infiniband machine.  It will be really helpful if this can be taken in
the cvs.


On Aug 15, 1:36 am, Axel <akoh... at> wrote:
> hi everybody,
> in the files section you'll find an updated patch relative to the
> current cvs (as of now ;-) ) that
> has some tweaking and a few required changes to make the code work at
> all with machines
> using the intel compiler where there is no option to have the default
> stack size raised (needed
> particularly for clusters based on redhat/centos/rocks/fedora/
> scientific linux/whitebox/etc.).
> it also removes one call to system in machine_intel.F that is
> interfering with most current
> infiniband installations. if you call system anytime on one of those
> machines, you may get
> unexpected crashes or hangs some time later when accessing the
> infiniband layer (not
> always right away!!) and disables an overloaded parallel call, that
> leads to crashes with
> many compilers and has not much use for a regular user (grid
> statistics).
> comments, bug reports etc. are welcome.
> i'm looking forward to have yet another discussion about the futility
> of adding c code to cp2k, educating sysadmins instead of working
> around their (well intended) paranoia and/or incompetence and need
> of getting rid of call system().
> ;)
> axel.
> ps: URL is

More information about the CP2K-user mailing list