A three-layer hierarchical approach for inventory control and production optimization of semiconductor reentrant manufacturing lines is developed. At the top layer, the parameters of an aggregated model are obtained online while, at the intermediate layer, production optimization and inventory control via model predictive control are performed. The aim of these two layers is aggregated (or averaged) supervisory control. The bottom layer consists of a distributed control policy which issues discrete-event decisions to track the aggregated targets issued by the optimizer. This layer accomplishes shop-floor control. The algorithm is applied to a discrete-event manufacturing line problem developed by Intel Corporation, which captures the main challenges posed by reentrant manufacturing lines.