An Ontology Based Methodology for Automated Algorithms Recognition in Source Code

Emerging computing architectures exploit parallel execution units for performances improvements in programs executions. Recent advances in microprocessors and multicomputers stressed the need of methodologies able to take advantage of parallelism. Source code analysis techniques exist which are used to identify chunks of code for parallelization purposes, but they are not designed to identify complex algorithmic patterns. In this work a methodology for identification of complex patterns is described. It is based on semantic description of patterns, which allows not only for reasoning, but also for providing an easy way to produce and extend algorithmic patterns definitions. A comparison of this approach with an attribute grammar language for patterns description is also reported.