Page 1 of 1

Incorrect mpi results on Itanium

Posted: Fri Feb 10, 2006 2:39 pm
by peterklaver
Hello everyone,

I'm trying to run VASP (4.6.26) on an Itanium-linux computer. The serial version seems to run fine (although I had to reduce the optimisation on radial.f90 to -O2), but the mpi version gives very wrong results for some calculations, even when everything is compiled with -O0. For other calculations the -O0 executable gives correct results.
Is anyone using the mpi version on Itanium-linux? Which compilers and BLAS versions are you using? I use ifort 9.0, mkl721, and the HP mpi in /contrib2/hpmpi. I've put input files for a testcase, along with my vasp.4.6 makefile and the correct and wrong OSZICAR output on

http://dutsm1078.stm.tudelft.nl/files.tar.gz

If someone is using the mpi version on Itanium, could you run these input files in parallel please and let me know the result?

kind greets,
Peter
t.p.c.klaver@qub.ac.uk

Incorrect mpi results on Itanium

Posted: Thu Feb 16, 2006 11:22 am
by peterklaver
Forget it, Martin Hilgeman from SGI helped me out already. It was some pesky little thing in mkl721 (undefined 'dlnu' internal mkl symbol). With mkl8 it works.

Peter