[CPMD-list] Wrong kinetic energy when running with LAM (but not for serial exec.)

[Axel Kohlmeyer]

On Tue, 20 Nov 2007, Henryk Wicke wrote:

HW> Dear all,

dear henryk,

HW> version which seems to be running fine so far. (Specific optimizations 
HW> used: -O2 -tpp7 -xT)

i found the vectorization options like -xT pretty dangerous
on many machines. the compilers seem much more likely to
miscompile parts of the code. furthermore, despite the claims
of compiler vendors SSE (or SIMD) vectorization does not bring
larger improvements (if any) for quantum chemistry packages,
since the SSE unit does not support much beyond a few elementary
opterations and then you have to switch back and forth between
the regular FPU and the SSE unit that this switching is frequently
so expensive that not using SSE can be faster. operations that
only use "pure" linear algebra in CPMD are mostly done via BLAS/LAPACK
and those libraries have optimized vectorization integrated 
(mostly by handcoded assembly!).

HW> For the parallel version I've also used LAM 7.1.4 which I had previously 
HW> built with icc/icpc 9.1.051 and ifort 9.1.043.

i would recommend switching to OpenMPI. it is pretty mature
these days and has better (=faster) implementations collective 
operations than LAM/MPI.

HW> The compilation of the parallel version went fine and first tests from 
HW> the test suite were promising. Now I went on to check some calculations 
HW> from the CPMD tutorial at A. Kohlmeyer's website and ran into trouble 
HW> with the calculation 2-h2o-pbc-geo-linsc in section 6.2.2. Especially 
HW> the kinetic energy in the "FINAL RESULTS" section is conspicuous...
HW> 
HW> The serial calculation seems to yield good results. The results differ 
HW> for the LAM calculations though:
HW> 
HW> Serial binary:
HW> KINETIC ENERGY =           12.82302639 A.U.
HW> 
HW> LAM with ONE CPU core:
HW> KINETIC ENERGY =           12.82302824 A.U.
HW> 
HW> Completely different for LAM with 2 cores:
HW> KINETIC ENERGY =            6.36582557 A.U.

this is the kinetic energy of the DFT part (not the MD).
how does the _total_ energy behave? this is the parameter
that matters. the individual components can fluctuate a bit.
(although half is not exactly a bit). 

cheers,
   axel.

HW> It might be interesting that the kinetic energy listed shortly before 
HW> the line "END OF GEOMETRY OPTIMIZATION" is still reasonable, i.e. 
HW> 12.82302... A.U., even in the last case.
HW> 

[...]

HW> Any help and/or insight is very much appreciated, especially because I'm 
HW> new at both CPMD as well as operating my own cluster and compiling such 
HW> relatively complex programs.



HW> 
HW> Regards,
HW> Henryk Wicke
HW> 
HW> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.