Performance comparisons of basic OpenMP constructs

OpenMP has become the de-facto standard for shared memory parallel programming. The directive based nature of OpenMP allows incremental and portable developement of parallel application for a wide range of platforms. The fact that OpenMP is easy to use implies that a lot of details are hidden from the end user. Therefore, basic factors like the runtime system, compiler optimizations and other implementation specific issues can have a significant impact on the performance of an OpenMP application. Frequently, OpenMP constructs can have widely varying performance on different operating platforms and even with different compilers on the same machine. This makes it very important to have a comparative study of the low-level performance of individual OpenMP constructs. In this paper, we present an enhanced set of microbenchmarks for OpenMP derived from the EPCC benchmarks and based on the SKaMPI benchmarking framework. We describe the methodology of evaluation followed by details of some of the constructs and their performance measurement. Results from experiments conducted on the IBM SP3 and the SUN SunFire systems are presented for each construct.