COMPASS: Community driven Parallelization

Parallel hardware1 has become a ubiquitous component in computer processing technology. Uniprocessors are slowly being phased out, with architectures evolving to allow greater performance gains [1] [2] [3] [4]. Exploiting the full potential of these architectures is a problem that has foxed programmers for a very long time. There have been several approaches to fully realize their potential. These include hardware based solutions, software based solutions and an amalgam of both. However, there is still a lot of scope in improving the performance outputs on these architectures. Needless to say no algorithm can successfully emulate the analytical skills of an experienced programmer. Inspired by this philosophy, this paper proposes ‘COMPASS ’ a novel ‘community driven approach’ whereby programmers are provided with a framework/platform for sharing programming expertise to improve performance of programs. One of the ways of increasing the usage of computational resources is using multiple threads of execution. A multi-threaded program is based on the premise that large problems can often be split into smaller independent problems which can be solved concurrently. Multi-threaded programs however, come with the increased overhead of handling concurrency issues (managing the thread life-cycle, efforts to avoid race conditions, deadlocks etc.). These overheads increase the design complexity, the logic and structure of the program manifold. To ease th