If specified VASP will use scaLAPACK instead of LAPACK
for the LU decomposition and diagonalization of the
sub space matrix (
).
These operations are very fast in
the serial version (
) but become a bottleneck on
massively parallel machine. If scaLAPACK is installed
on the parallel machine use this switch. scaLAPACK can be used on
the T3E only for programming enviromente 3.0.1.0. (3.0.0.0 does for
instance not offer the required routines). On the
T3D (but not T3E) the additional switch
-DT3D_SCAmust be specified, at least for the scaLAPACK version tested by us (the T3D scalapack is not compatible to standard scaLAPACK routines).
scaLAPACK can be switched of during runtime by specifying
LSCALAPACK = .FALSE.in the INCAR file. (this is required in the nudged chain mode on the T3D, because currently scaLAPACK on the T3D is not supporting the routine BLACS_GRIDMAP, see Sec. 8.4).
MIND: in the Gamma point only version, the parallel sub space diagonalization (LscaLAPACK= True) is performed with a Jacobi algorithm instead of scaLAPACK. This routine was written by Ian Bush. The Jacobi routine is faster than scaLAPACK.