MAP-i Doctoral Programme Parallel Programming by Transformation

The complexity of hardware platforms available today is increasing. Com-plex memory hierarchies, multi-core processors, GPUs as coprocessors, etc. bring new challenges to software developers. Developing efficient applica-tions to a specific platform becomes increasingly more difficult. Additionally, developers usually have to maintain versions optimized to different platforms available, and it is usually difficult to the average programmer to understand highly optimized software. This research addresses these challenges using model-driven engineering, a programming methodology that deals with software complexity raising the level of abstraction in program development. As a first step, we will focus on a specific area of application, Dense Linear Algebra, with the goal to automate development of customized, but well-understood, dense linear algebra programs. At the same time, we expect that our approach will not be domain-specific. Work in other domains will help confirm the generalit