GPU accelerator ; symmetric tridiagonal reduction ; eigenvalue problems ; hybrid architectures ; Householder transformation ; one-stage ; two-stage ; ridiagonal

Solutions of eigenvalues and eigenvectors of complex Hermitian matrices are widespread and have a very important role in scientific calculations. These solutions can be obtained from linear algebra libraries by the functions of Lapack and by its parallel version ScaLapack in systems with general and distributed memory, respectively. However, it’s more beneficial to get these solutions in hybrid architectures which require a new development of algorithms to efficiently organize non-uniformity and massive parallelization in a graphical processor. The main objective of this paper is to present the performance of standard, as well as generalized case of solutions of eigenvalues and eigenvectors of complex and double complex Hermitian matrices on Tesla C1060 accelerator.