Synthesis Oriented Scheduling of Multiparty Rendezvous in Transaction Level Models

Abstract

Due to the large semantic gap between transaction level models and actual implementations, hardware synthesis based on system level models has been a great challenge. Aiming to close the semantic gap, we studied an approach that uses rendezvous to model communication. By allowing both conjunctive and disjunctive composition of rendezvous, the approach supports flexible communication patterns involving multiple processes. However, a practical issue of the model is the complexity of scheduling of multiparty rendezvous, which is NP hard in general. This paper proposes an efficient scheduling algorithm. It begins by encapsulating state transition information of processes into a relation graph. It then creates a tree that relates edge combinations. The tree is used to guide the scheduler at run time to search for schedulable sets. Experimental results prove that this algorithm improves our scheduler significantly. The algorithm lays the ground for the synthesis of the communication and synchronization circuitry for the system.