This article considers the design of an anti-windup compensator for linear systems subject to time-varying input delays and saturating actuators. Local and global stabilization conditions ensuring both external as well as internal stability of the closed-loop system are derived directly as linear matrix inequalities (LMIs). To compute the anti-windup gains, these conditions are cast into the following optimization problems: maximization of the set of admissible initial conditions, maximization of the bound of admissible L2 disturbances or the maximization of the L2-gain from the disturbance to the regulated output. Simulation examples are provided to illustrate the proposed solution.