Pattern Candidate Discovery and Parallelization Techniques

Abstract

Parallel computations in a program can be expressed conveniently, at a high level of abstraction, using parallel patterns such as task farm, pipeline or divide-and-conquer. In order to transform a sequential program into a pattern-based parallel one, the software developer may want to apply refactoring transformations on it. This tutorial explains a methodology to perform tool supported parallelization of programs by presenting how to use a specific static source code analysis and transformation system for Erlang. A key element of the approach is pattern candidate discovery, a static analysis technique to identify code fragments that can be refactored into a parallel pattern.