OpenMP and HPF: Integrating Two Paradigms

Recently a proposal was put forth for a set of language extensions to Fortran and C based upon a fork-join model of parallel execution; called OpenMP, it aims to provide a portable shared memory programming interface for shared memory and low latency systems. Current distributed memory systems are increasingly configured with nodes consisting of multiple processors which share memory. For these, both shared memory and distributed memory programming paradigms offer functionality appropriate for application development.

We discuss both the HPF and the OpenMP programming models, and consider how they may be used together to write programs which exploit the full capabilities of such systems. We show that the HPF extrinsic mechanism provides for several simple interfaces which can satisfy this likely need. Given the orthogonality of some of the concepts in the two programming models, there appears to be added value in providing some level of integration of the two. In particular, OpenMP could benefit from features to specify data locality. Since at least one major vendor already has provided some data mapping options to a shared memory parallel programming model [1], we expect that this avenue will be closely explored by the OpenMP consortium in the near future.

References