[DFTB-Plus-User] compile DFTB+ for AMD opteron processors
Bálint Aradi
balint.aradi at bccms.uni-bremen.de
Tue Apr 12 11:59:12 CEST 2011
Dear Chen,
> I have complied the DFTB+ source code using intel compiler in my linux server. My linux server has 32 AMD Opteron processors. However, the efficency is lower than the server with 8 Intel processors. Could you give me some advices for improving the efficency of this architecture?
You should definitely not use the original F77 Lapack and BLAS routines
(libblas.a and liblapack.a) as they are much slower, as the optimized
MKL routines. I do not know, however, how MKL behaves on AMD machines,
they were many issues a few years ago, MKL being slower on purpose on
AMD machines. You could try to compile ATLAS for your system. You would
need, however, needed the threaded version if you intend to run on more
than one core at the same time. I had some difficulties with that in the
past, maybe it works now properly.
>
> I have also tried using PGI compiler for this architecture. But while runing make install, there was a problem:
>
> make[1]: *** [type_geometry_hsd.o] Error 2
> rm md_integrator.f90 slako_eq_grid.f90 short_gamma.f90 spin.f90 geoopt.f90 disp_common.f90 tokenreader.f90 intrinsicpr.f90 formatout.f90 bisection.f90 inputdata_.f90 fifo.f90 ranlux.f90 rep_poly.f90 memman.f90 diis_mixer.f90 sort.f90 md_common.f90 coulomb.f90 type_geometry.f90 shift.f90 dftb_pls_u.f90 eigensolver.f90 andersen_thermostat.f90 extcharge.f90 dispersion.f90 orbital_equiv.f90 xmlutils.f90 qm.f90 sk.f90 wrapped_pointers.f90 ext_erfc.f90 linmin.f90 accuracy.f90 mixer_adt.f90 general_list.f90 disp_uff_data.f90 hsdparser.f90 lapackroutines.f90 eigenvects.f90 type_geometry_hsd.f90 linkedlist.f90 velocity_verlet.f90 charmanip.f90 hsdutils.f90 energies.f90 scc.f90 disp_slaterkirkwood.f90 unitconversion.f90 potentials.f90 broyden_mixer.f90 mainio.f90 message.f90 blasroutines.f90 thermostat.f90 nonscc.f90 numericalDerivs.f90 populations.f90 periodic.f90 hsdutils2.f90 temp_profile.f90 oldcompat.f90 slako_cont.f90 charge_constraints.f90 berendsen_thermostat.f90 simple_algeb
ra.f90 initprogram.f90 stringlist.f90 rep_spline.f90 disp_uff.f90 common_types.f90 dummy_thermostat.f90 conjgrad.f90 sparse2dense.f90 scc_init.f90 simple_mixer.f90 stepdesc.f90 rep_cont.f90 allocate.f90 fileid.f90 angmomentum.f90 constants.f90 interpolation.f90 anderson_mixer.f90
> make[1]: Leaving directory `/home/hycheng/opt/dftb+_1.1_src/prg_dftb/_obj_x86_64-linux-pgi'
> make: *** [_obj_x86_64-linux-pgi] Error 2
You did not send the actual lines with the error message (which come
before the lines starting with ***-). Could you send those as well?
But in terms of Fortran 95 compatibility, PGI was the worst compiler
among those I tried ca. 2-3 years ago, not being able to compile DFTB+
at that time at all. Since then I do not have access to PGI compilers
any more, but probably there are still issues.
Best regards,
Bálint
--
Dr. Bálint Aradi
Bremen Center for Computational Materials Science, University of Bremen
http://www.bccms.uni-bremen.de/en/~b_aradi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 262 bytes
Desc: OpenPGP digital signature
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20110412/2e13ea5b/attachment.sig>
More information about the DFTB-Plus-User
mailing list