Abstract
In this paper we describe a novel approach to the scheduling and instruction selection phases of code generation. Our approach mimics that used by human programmers and combines heuristic search with hierarchical planning and a modified means-ends analysis. Optimal code generation is known to be NP-Complete hard. Our algorithm can be executed in time polynomial to the size of the input program and exponential only to the depth of the search. Empirical results show that very good results can be accomplished with depths as small as 3. The quality of the generated code is comparable (in some cases superior) to that of codes generated by human digital signal processing experts.
Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
