Efficient Language-Based Parallelization of Computational Problems Using Cilk Plus

The aim of this paper is to evaluate Cilk Plus as a language-based tool for simple and efficient parallelization of recursively defined computational problems and other problems that need both task and data parallelization techniques. We show that existing source codes can be easily transformed to programs that can utilize multiple cores and additionally offload some computations to coprocessors like Intel Xeon Phi. We also advise how to improve simplicity and performance of data parallel algorithms by tuning data structures to utilize vector extensions of modern processors. Numerical experiments show that in most cases our Cilk Plus versions of Adaptive Simpson’s Integration and Belman-Ford Algorithm for solving single-source shortest-path problems achieve better performance than corresponding OpenMP programs.