Mixed-Precision Orthogonalization Scheme and Adaptive Step Size for CA-GMRES on GPUsдоклад на конференции