Knowledge-Based Code Selection Methods in Retargetable Microcode Synthesis

Abstract

We have presented a method for machine-independent synthesis of microcode using knowledge-based techniques. A high-level representation of the microprogram is translated into symbolic assertions and then converted by the synthesis system to microcode. Thus, a synthesis system can function as the back end of a microcode compiler. Since the knowledge base is machine-independent and requires only a machine description to generate microcode for that machine, it can also be used as a retargetable microcode compiler. The knowledge base is in procedural form and consists of transformation rules that reflect the effect of MOs on machine states. Many of these rules simply restrict the transformations to acceptable behavior on the part of the synthesis system. Others help to reduce the search by providing knowledge about desirable behavior in the form of heuristics, or help the synthesis system reason about different possible derivations for the same specification.