Hi Dan..
> I just downloaded the latest version of intel's free compiler suite for
> linux, which comes with the latest version of MKL (10.2). Currently the
> configure_at script in amber11 does not deal with this version
> correctly,
> resulting in linking failures. After some poking around in the MKL
> manual (
> http://software.intel.com/sites/products/documentation/hpc/mkl/lin/inde
> x.htm)
> it seems the correct way to link should be:
<sigh>. More examples of the patients running the asylum!!!
Note we need to be even more careful than this. There are two ways to link
MKL - the serial way and the openmp parallel way. What you show below is the
openmp parallel way I think. We want to avoid this at all costs because it
causes big problems since they went and openmp paralleled the whole vector
math library etc. Disaster if you are running mpi and forget to set
OMP_NUM_THREADS=1 on all nodes. Thus by default we should be linking against
the serial version of MKL. However we want to preserve the ability to link
against the parallel openmp version since one can specify -openmp to
configure and then it will use the parallel diagonalizer in MKL for QMMM.
There is protection in the openmp code to force runtime OMP_NUM_THREADS to 1
(using an openMP call) for all the other MKL routines, like vdinvsqrt.
All the best
Ross
/\
\/
|\oss Walker
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
|
http://www.rosswalker.co.uk | PGP Key available on request |
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Wed Aug 19 2009 - 22:17:06 PDT
