Blocked algorithms and software for reduction of a regular matrix pair to generalized Schur form

Abstract

A two-stage blocked algorithm for reduction of a regular matrix pair ( A , B ) to upper Hessenberg-triangular form is presented. In stage 1 ( A, B is reduced to block upper Hessenberg-triangular form using mainly level 3 (matrix-matrix) operations that permit data reuse in the higher levels of a memory hierarchy. In the second stage all but one of the r subdiagonals of the block Hessenberg A -part are set to zero using Givens rotations. The algorithm proceeds in a sequence of supersweeps, each reducing m columns. The updates with respect to row and column rotations are organized to reference consecutive columns of A and B . To further improve the data locality, all rotations produced in a supersweep are stored to enable a left-looking reference pattern, i.e., all updates are delayed until they are required for the continuation of the supersweep. Moreover, we present a blocked variant of the single-diagonal double-shift QZ method for computing the generalized Schur form of ( A, B in upper Hessenberg-triangular form. The blocking for improved data locality is done similarly, now by restructuring the reference pattern of the updates associated with the bulge chasing in the QZ iteration. Timing results show that our new blocked variants outperform the current LAPACK routines, including drivers for the generalized eigenvalue problem, by a factor 2-5 for sufficiently large problems.