The Structure of a Compiler for Explicit and Implicit Parallelism

We describe the structure of a compilation system that generates code for processor architectures supporting both explicit and implicit parallel threads. Such architectures are small extensions of recently proposed speculative processors. They can extract parallelism speculatively from a sequential instruction stream (implicit threading) and they can execute explicit parallel code sections as a multiprocessor (explicit threading). Although the feasibility of such mixed execution modes is often tacitly assumed in the discussion of speculative execution schemes, little experience exists about their performance and compilation issues. In prior work we have proposed the Multiplex architecture [1], supporting such a scheme. The present paper describes the compilation system of Multiplex. Our compilation system integrates the Polaris preprocessor with the Gnu C code generating compiler. We describe the major components that are involved in generating explicit and implicit threads. We describe in more detail two components that represent significant open issues. The first issue is the integration of the parallelizing preprocessor with the code generator. The second issue is the decision when to generate explicit and when to generate implicit threads. Our compilation process is fully automated.