Direct energy deposition (DED) is a metal additive process that has applications in high-value part repair and fabrication of functionally-graded parts. However, the process is sensitive to commanded inputs and process conditions, such as powder flow and heat conduction from the melt pool, which change throughout the course of a build and often lead to geometric inaccuracies in the final part. Thus, there is a need for in-process sensing and feedback control to improve robustness to process conditions and achieve the desired part geometry. Previously, a repetitive process, quadratic-optimal height controller was implemented on thin-wall builds, where height measurements and control updates were performed in between layers. In that work, the desired layer thickness remained constant from layer to layer. Here, the repetitive process framework is extended to account for height references that change with layer number, as will be the case for producing more complex part geometries. Using this extended model, a model predictive controller is derived and simulated on a part build with iteration-varying references.