I am using LAPACK routines in order to determine eigenvalues
of symmetric or hermitean matrices. These routines are contained
in the MKL and in principle allow threads. Looking at the
perfomance I realize that they are not or only poorly parallelized.

Does anybody know whether one of the routines
ZHEEV, ZHEEVD, or ZHEEVR or a similar one
has been worked over to use multiple threads for
matrix diagonalization?

Thanks a lot, Juergen Schnack (Osnabrueck, Germany)