A Study on OpenMP and Irregular Programming

Abstract

Programmers need better, more reliable techniques to build parallel programs. In the task-parallel programming paradigm, the programmer defines independent pieces of work, called tasks that may run in parallel. Geometric algorithms are wonderful examples for programming, because they are deceptively easy to do with your eyes, yet much harder to implement for a machine. The goal of parallel algorithm design is to develop parallel computational methods that run very fast with as few processors as possible. OpenMP(Open Multi-Processing) uses shared memory architecture to implement parallel programming in dual core machines. A proper implementation expected to give a speedup around 1.5 in dual core machine and 2.8 in a quad core machine. Writing irregular parallel algorithms with OpenMP has been rarely practiced in the past. Yet it is possible, and in this paper we will study the implementation of irregular parallel geometry algorithms to show a possible stepping stone and deficiency of OpenMP specification.