Performance Evaluation of OpenMP and CUDA on Multicore Systems

摘要

Nowadays, not only CPU but also GPU goes along the trend of multi-core processors. Parallel processing presents not only an opportunity but also a challenge at the same time. To explicitly parallelize the software by programmers or compilers is the key for enhancing the performance on multi-core chip. In this paper, we first introduce some of the automatic parallel tools based OpenMP, which could save the time to rewrite codes for parallel processing on multicore system. Then we focus on ROSE and explore it in depth. And we also implement an interface to reduce its complexity of use and use some automatic parallelization for CUDA.