Semantic analysis for parallelizing C

Abstract

Automatic parallelization of C language programs is a difficult task due, for example, to side effects within expressions and access to memory via pointers. We describe a method based on semantic analysis of the language, in order to detect parallelism in complex programs using pointer arithmetic. We define semantic domains which respect the type system of the language, and use them to synthesize effect of the program through functional objects called transformations.