Phase Structure of SUGAR

The preceding sections described the important phases in the multiphase method used by SUGAR to synthesize microprocessors. Descriptions of phases which were not described can be found in [RajanSS]. This section draws on the earlier descriptions to show how the phases are organized and how they communicate with each other. 

Each phase solves one or more subproblems in synthesis. The phases communicate with each other through internal behavioral descriptions which are either TOOL syntax trees or data flow graphs. Each phase transforms, annotates or translates its input. LISP property lists, the primary communication mechanism between phases, are used to record descriptions of design decisions and design analyses. Information recorded by earlier phases is kept as long as it is needed by later phases. 

A delayed binding strategy is the primary method used to resolve interactions between phases. Decisions are delayed until enough information is available to make them. Some phases are repeated so that conservative design decisions made in an earlier pass can be 

While sugar s complex multi-phase strategy has proved to be useful in resolving some difficult interactions in microprocessor synthesis, additional work is necessary to resolve interactions which SUGAR does not address. Of these, the most serious interactions are those between the transformation phases and the synthesis phases. 

SUGAR is written in FRANZ LISP, a dialect of LISP, and consists of more than 35 phases and over 20,000 lines of code. Analysis of the performance of the program is difficult since it contains many algorithms and many rules. 

---