OpenMP tasking analysis for programers

As of 2008, the OpenMP 3.0 standard includes task support allowing programmers to exploit irregular parallelism. Although several compilers are providing support for this new feature there has not been extensive investigation into the real possibilities of this extension. Several papers have discussed the programming model itself while other papers have discussed design and implementation on different platforms. There are also papers demonstrating performance results using well known kernel applications. This paper presents an analysis of the OpenMP tasking model possibilities, using the IBM XL compiler implementation. Using different parameters such as the number of tasks, task granularity and parallelism pattern, this paper explores how such parameters can affect the average performance and identifies the limits of the OpenMP tasking model.Peer ReviewedPostprint (published version