MPI-GPU parallelism in iterative eigensolvers for block-tridiagonal matrices