This paper proposes power allocation policies for the secondary users (SUs) of a cognitive radio network (CRN). The SUs transmit in the same band as the primary user (PU) where the latter employs Type-II Hybrid-ARQ with incremental redundancy (IR-HARQ). Each SU receiver applies interference cancelation to remove a successfully decoded PU message in the subsequent PU retransmissions of the same message. The SU transmitters exploit this PU message knowledge state at their receivers as well as the accumulated mutual information (ACMI) and ACK/NACK feedback transmitted by the PU receiver. The ACK/NACK message demonstrates the successful/unsuccessful decoding of the PU message at the PU receiver. With this approach and using a Constrained Markov Decision Process (CMDP) model, centralized optimum power adaptation policies for the SUs are proposed to maximize their average sum throughput under constraints on the PU throughput and on the maximum and average SU transmission powers.