Date: April 11, 1996

Speaker: Jaspal Subhlok

HPF Task Parallelism

Abstract:
Task parallelism can play an important role in a data parallel language like High Performance Fortran. Motivation for task parallelism are both expressiveness and performance. Several applications can be programmed more naturally using a combination of task and data parallelism, and several other applications can use task parallelism to improve performance and scalability. I have been working with the HPF Forum to develop task parallelism for HPF and finally there is broad consensus on the approach. I will present the task parallelism model for HPF 2.0 which is is of the same flavor asthe Fx task parallelism model. It emphasises simplicity, sequential consistency and ease of compilation, but trades off some power of expression in the process.

SDI / LCS Seminar Questions?
Karen Lindenfelser, 86716, or visit www.pdl.cmu.edu/SDI/