Parallel one-pass compilation

Abstract

The parallelization of one-pass compilers is studied systematically. The approach is not restricted to different features in certain languages; instead, L-attributed grammars are used as the model of compilation. This language-independent approach can be used to generate a parallel compiler on the basis of a formal specification of the source language. The scheduling strategy is based on a static examination of the attribute dependencies of a production rule. Thus the potential of parallelism at runtime depends on the attribute dependencies of the grammar. Parallel (asynchronous) processes communicating via events serve as the machine model for attribute evaluation. Aiming at a full model of one-pass compilation, the pure L-attributed model is extended with pipelined parsing. In order to gain experience with parallel attribute evaluation, a simulator has been implemented. Experiments with the simulating system are described and interpreted.