Irregular Parallelism Challenge Sample Clauses

Irregular Parallelism Challenge. 1. A first detailed analysis of the performance of three widely-used threading li- braries on a highly optimized program with high levels of irregular and unbal- anced tasks on the Xeon Phi is provided. 2. A straightforward First In, First Out (FIFO) scheduling policy is shown to be equal or even to outperform the more elaborate threading libraries Cilk Plus and Threading Building Blocks (TBB) for running high levels of parallelism for high numbers of cores. This is surprising since Cilk Plus was designed to achieve high efficiency for precisely these types of applications. 3. The first parallel MCTS with grain size control is proposed. It achieves, to the best of our knowledge, the fastest implementation of a parallel MCTS on the 61-core Xeon Phi 7120P (using a real application) with a 47 times speedup compared to sequential execution on the Xeon Phi itself (which translates to 5.6 times speedup compared to the sequential version on the regular host CPU, Xeon E5-2596).