Criteria-Based Merging of Dynamic CAL Actors

The dataflow model of computation is well-established in many application areas, including image encoding and decoding. This model consists of actors that process data and communication channels that transmit data between actors. However, when mapping these models to the systems on which they will be executed and synthesizing the necessary code, the number of actors often exceeds the number of execution units by a significant amount. This results in increased overhead for data buffering and communication between actors. To address this issue and adapt the model to the target execution system, we propose a technique that combines static and dynamic actors defined in the CAL actor language. This technique is based on actors’ composition within the dataflow network and creates a new composite CAL actor. When merging actors, we consider the actions of the actors, including their CAL scheduling mechanisms, finite state machines, priorities, and guards. We derive conditions under which connected actors can be merged to form a new CAL-compliant actor. Preliminary experiments show performance improvements of up to 3.6x compared to unoptimized code.