VHQRII - AN ACCURATE, MODULAR AND FAST DIAGONALIZATION ROUTINE FOR VECTOR PROCESSORS

Based on subroutine HQRII1, the fastest eigensolver for real dense symmetric matrices, the authors developed a modular version of it. All N**3-dependent steps involving floating point operations are expressed as calls to inner product and vector-scalar-multiply-and-add modules which may easily be optimized to exploit a vector architecture. An implementation which uses the vector library of HP-1000 series computers is discussed in detail.