akoh... at gmail.com
Tue Jul 15 16:39:48 CEST 2008
On Jul 14, 3:09 pm, Eric Shamay <eric.... at gmail.com> wrote:
> Thanks Axel,
> I've decided to recompile using gfortran and the gnu C compiler (gcc)
> instead of the Intel based ones for now. The compile was done from a
> fresh update from CVS. Now the reference errors are gone, but the
> compilation hangs while trying to put together qs_mo_types.F. I
> managed to compile quite successfully and repeatedly on an ia64
> machine about 6 months ago, but the recent CVS updates have somehow
> broken that by either hanging during compilation, or causing new
> undefined references.
> I'll keep hacking at it, but I'm still lost as to why the parallel
> compiles have stopped working altogether.
i have not had problems getting cp2k compiled in parallel over a long
time, except with broken compilers.
the compiler "hangs" are usually the result of either a compiler bug
or the compiler trying to optimize something that is difficult to
or using an algorithm for optimization that doesn't scale well with
complexity of the problem. this frequently happens in code that is not
at all time critical. files like f77_interface.F or pint_types.F are
typical examples. for example the IBM xlf compiler is quite sensitive
to this (it tries _very_ hard to optimize if you use a high
level), but the situation can be easily alleviated by reducing the
optimization level for those specific files. see for example the
Linux-ppc5-xlf.popt arch file.
More information about the CP2K-user