The generic real double and single precision kernels are basically
identical, thus one can use the same template for them.

Since Fortran lacks proper templates, the template is produced
with preprocessor defines

A random matrix is set up, and then result = a**T * 2*b is computed
this is compared with the result of plain scalapack routines and
the maximum norm of "scalapack - elpa" is computed (should be zero on
machine precision)