A constructive method for exploiting code motion

In this paper we address a resource-constrained optimization problem for behavioral descriptions containing conditionals. In high-level synthesis of ASICs or in code generation for ASIPs, most methods use greedy choices in such a way that the search space is limited by the applied heuristics. For example, they might miss opportunities to optimize across basic block boundaries when treating conditional execution. We propose an approach based on local search and present a constructive method to allow unrestricted types of code motion, while keeping optimal solutions in the search space. A code-motion pruning technique is presented for cost functions optimizing schedule lengths. A technique for treating concurrent flows of execution is also described.