With the aim of boosting the security of the conventional directional modulation (DM) network, a secure DM network assisted by intelligent reflecting surface (IRS) is investigated in this paper. To maximize the security rate (SR), we jointly optimize the power allocation (PA) factor, confidential message (CM) beamforming, artificial noise (AN) beamforming, and IRS reflected beamforming. To tackle the formulated problem, a maximizing SR with high-performance (Max-SR-HP) scheme is proposed, where the PA factor, CM beamforming, AN beamforming, and IRS phase shift matrix are derived by the derivative operation, generalized Rayleigh-Rize, generalized power iteration, and semidefinite relaxation criteria, respectively. Given that the high complexity of the above scheme, a maximizing SR with low-complexity (Max-SR-LC) scheme is proposed, which employs the generalized leakage and successive convex approximation algorithms to derive the variables. Simulation results show that both the proposed schemes can significantly boost the SR performance, and are better than the equal PA, no IRS and random phase shift IRS schemes.