Multithreaded parallelism with openmp

While multithreaded programming is an effective way to exploit concurrency, multithreaded programs are notoriously hard to program, debug and tune for performance. In this paper, we present OpenMP shared memory programming as a viable alternative and a much simpler way to write multithreaded programs. We show through empirical results obtained by running, on a single processor machine, a simple matrix multiplication program written in OpenMP C that the drop in performance compared with the single threaded version even on a uniprocessor machine may be negligible. However, this is well compensated for by the increased programmer productivity resulting from the ease of programming, debugging, tuning and the relative ease of OpenMP skill acquisition.