Optimization of symmetric eigendecomposition

Submitted by Tadej Ciglarič

Assigned to Nobody

Link to original bugzilla bug (#1763)
Version: 3.4 (development)

Description

I implemented a symmetric eigendecomposition algorithm that is significantly faster than one currently in Eigen (~3x on Core i5 2500). I achieved that speedup by using blocked tridiagonalization and replaced QR iteration with MRRR.

If you are interested I would like to add that to Eigen. Currently my code works with symmetric matrices of doubles, but I think it should not be too much work to adapt it for general selfadjoint matrices.

Blocking

#1608