A symbolic approach for the combined solution of scheduling and allocation

Abstract

Scheduling is widely recognized as a very important step in high-level synthesis. Nevertheless, it is usually done without taking into account the effects on the actual hardware implementation. This paper presents an efficient symbolic technique to concurrently integrate operation scheduling and resource allocation. The technique inherits all the features of "standard" BDD-based control dominated scheduling, including resource-constraining, speculation and pruning. In addition, it introduces an efficient way of encoding allocation information within a symbolic scheduling automaton with a two-folded target. Firstly, it finds a minimum cost allocation of operation resources satisfying a given schedule. Secondly, it optimizes the amount of registers required to store intermediate results of operations. Theory and algorithms are developed and presented. Experimental results on a well known set of benchmarks show the potentiality of the approach.