CP2K in IBM PWR5 machinces

[Axel]

dear yunfeng!

On Mar 17, 12:15 pm, "LIANG Yunfeng" <liangy... at gmail.com> wrote:
> Dear Fawzi,
>
> I have now compiled a working cp2k in AMD Opertron machince. It is
> encouraging...

excellent!

> Now I'm convinced that I have to compile lapack by my own. But I found it is
> related to lots of flags...could you attach your make.inc file (for lapack)
> for me?

you don't really need a lot of flags
(just use the ones that cp2k suggests and
perhaps lower the optimization level).
the major performance from LAPACK comes
through the use of BLAS and that one is
contained in an optimized form in ESSL.


> Second question is: in IBM SMP machince, does cp2k work with BLAS (netlib)

you don't need to compile BLAS. just use ESSL instead.
the main reason for compiling LAPACK on your own is that
IBM's ESSL contains a few LAPACK calls where the calling
sequence is different from the current LAPACK release.
by compiling LAPACK yourself (and making sure that this
library is referenced before ESSL) you use the standard
LAPACK subroutine instead of the IBM version.

> as well? If works, what should I be aware? I made one through BLAS/LAPACK
> and FFTW3, but I met the same problem as reported before. Maybe it is also
> flag problem...
>
> Regarding to ESSL, should and could I compile one by myself? Because I met
> the same problem as reported before /or even worse---just failed in

ESSL is provided as a binary by IBM (IBM pays people to optimize
it next to perfect for their machines).

> compiling cp2k, when I compiled the lapack-3.1.1 with ESSL. I know it may be
> related to the first question too.


hmmm.... you have to post the arch file and the error message so
that we can make sense of it. there are quite a few things that
can go wrong and sometimes the ibm linker is giving quite strange
error messages or warnings.

cheers,
   axel.
>
> Thanks
> Sincerely, Yunfeng
>
> On 3/8/08, Fawzi Mohamed <fa... at gmx.ch> wrote:
>
>
>
> > Hi Yunfeng,
> > It looks like a library problem (but it could also be the compiler).
> > You can contact the administrator of the machine to see if he is aware of
> > something, and if he can help you out (maybe you are using the wrong version
> > of the library, or he didn't update the libraries with the new compiler, but
> > that was needed).
> > Then as Teo suggested you can try to compile the lapack routines from
> >http://www.netlib.org/lapack/and test them (with their test program).
> > Then compile using your lapack version.
> > And finally file some bug reports to IBM, the more the problem is clear,
> > the more they will like it, hopefully the faster they will fix it.
>
> > sorry that your experience with cp2k was difficult, but there is hope,
> > cp2k can work (and well) on PWR5 ;)
>
> > ciao
> > Fawzi
> > On Mar 7, 2008, at 7:58 PM, Teodoro Laino wrote:
>
> > ok.. the AIX machine we use has the following compiler:
> > IBM XL Fortran Advanced Edition V10.1 for Linux
> > Version: 10.01.0000.0003
>
> > Since that input file (H2O.inp) is known to work, your issue is related
> > very probably to the libraries..
> > At least for the serial version you can copmiler blas/lapack by yourself
> > and check that the issue is not
> > the compiler..
>
> > cheers
> > teo
>
> > On 7 Mar 2008, at 19:34, LIANG Yunfeng wrote:
>
> > Which input file of the test directory were you running?
> > Can you run the tests/QS/H2O.inp ?
>
> > I was using exactly the tests/QS/H2O.inp, simply moving it into another
> > directory...
>
> > Sincerely, Yunfeng
>
> > On Fri, Mar 7, 2008 at 12:05 PM, Teodoro Laino <teodor... at gmail.com>
> > wrote:
>
> > > Dear Yunfeng,
> > > Which input file of the test directory were you running?
> > > Can you run the tests/QS/H2O.inp ?
>
> > > Anyway this error message means that the problem is in your library..
> > > Compiler version (xlf95_r -qversion)?
>
> > > Teo
>
> > >  On 7 Mar 2008, at 19:00, LIANG Yunfeng wrote:
>
> > >  Then I tested it in water monomer case, again using the same input
> > > copied from /cp2k/tests/QS/
>
> > > the program stops at:
>
> > > .....
> > >  SCF WAVEFUNCTION OPTIMIZATION
> > >   Step  Update method              Time         Convergence        Total
> > > energy
>
> > > -----------------------------------------------------------------------------
>
> > >   ----------------------------------- OT
> > > --------------------------------------
>
> > >   Allowing for rotations:  F
> > >   minimizer      : DIIS                : direct inversion
> > >                                          in the iterative subspace
> > >                             using      : -   7 diis vectors
> > >                                          - safer DIIS on
> > >   preconditioner : FULL_ALL            : diagonalization, state
> > > selective
> > >   stepsize       :    0.15000000
> > >   energy_gap     :    0.00100000
> > >   eps_taylor     :   0.10000E-15
> > >   max_taylor     :             4
>
> > >   ----------------------------------- OT
> > > --------------------------------------
>
> > >  *
> > >  *** ERROR in cp_fm_syevd (MODULE cp_fm_diag)  ***
> > >  *
>
> > >  *** Matrix diagonalization failed  ***
>
> > >  *** Program stopped at line number 151 of MODULE cp_fm_diag  ***
> > >  ===== Routine Calling Stack =====
> > >             7 cp_fm_syevd
> > >             6 subspace_eigenvalues_ks
> > >             5 init_scf_loop
> > >             4 scf_env_do_scf
> > >             3 qs_energies
> > >             2 qs_forces
> > >             1 CP2K
> > >  CP2K| Abnormal program termination, stopped by process number 0