When integrating device-to-device (D2D) communications with densely deployed cellular networks, both energy efficiency (EE) and quality of service (QoS) will be severely degraded by strong intracell and intercell interference. To optimize EE while guaranteeing QoS provisioning, a three-stage energy-efficient resource allocation algorithm is proposed, which combines centralized interference mitigation and distributed power allocation algorithms by exploiting multi-cell cooperations, nonco-operative game, nonlinear fractional programming, and Lagrange dual decomposition. Simulation results have demonstrated that the proposed algorithm achieves a nearly zero infeasibility ratio, and improves EE performance significantly for both cellular and D2D user equipments (UEs) compared to the previous distributed scheme.