Instruction-set-extension exploration using decomposable heuristic search

Instruction-set-extensions (ISEs) are custom instructions designed specifically for a class of applications. Hence extracting the maximum benefit out of them is any designer's goal. ISE application is the problem of efficiently applying custom instructions such that some gain function like code-size is optimized. Traditional algorithmic techniques like ILP and state space search have failed to address practical designs. Hence a considerable effort has been put on the design of heuristic algorithms. In this paper we present a new optimal algorithm DCA* that combines divide-and-conquer technique with heuristic search to address this problem. We demonstrate the effectiveness of this novel algorithm on real life designs.