EXPERIENCES WITH ASPECT-BASED PARALLELIZATION OF SCIENTIFIC CODE

This paper discusses the use of Aspect-Oriented Programming (AOP) to support the parallelization of scientific code. The idea is to develop parallelism concerns in separate aspects so that the weaving process can inject the code structures which allow the sequential scientific core to be executed in parallel. A series of advantages can initially be derived from this aspect-based approach. As parallelism is modularized into separate units, the code-tangling level is reduced. The parallel version is developed by reusing the code pieces which implement the numerical computation sequentially. Moreover, different aspects can be written to adapt the application to different high-performance environments. This paper describes some experiences with the code parallelization using aspects. Specifically, the integration of both task and data parallelism in the context of two parallel scenarios (i.e. multithreading and message-passing) is addressed in a case study. The work is an attempt to assess the benefits and limitations of applying AOP for these purposes