A Data Parallel Formulation of the Barnes-Hut Method for N-Body Simulations
This paper presents a data-parallel formulation for N-body simulations using the Barnes-Hut method. The tree-structured problem is first linearized by using space-filling curves. This process allows us to use standard data distributions and parallel array operations available in data-parallel languages. A new efficient HPF implementation of the Barnes-Hut method is presented in this paper, characterized by the use of array copy sections to express communications, obtaining efficient point- to-point data transferences. In addition, HPF_LOCAL constructs are used to obtain very compact and efficient local (node) codes.
KeywordsTree Data Structure Force Contribution Owned Body High Performance Fortran Irregular Problem
Unable to display preview. Download preview PDF.
- 3.Hu, Y.C., Johnsson, S.L., Teng, S.-H.: High Performance Fortran for Highly Irregular Problems. ACM SIGPLAN Symp. Principles and Practice of Parallel Programming (PPoPP’97) (Las Vegas, NV, Jun. 1997) 13–24.Google Scholar
- 4.McCurdy, C., Mellor-Crummey, J.: An Evaluation of Computing Paradigms for N-body Simulations on Distributed Memory Architectures. ACM SIGPLAN Symp. Principles and Practice of Parallel Programming (PPoPP’99) (Atlanta, GA, May 1999) 25–36.Google Scholar
- 5.Ou, C., Ranka, S., Fox, G.: Fast and Parallel Mapping Algorithms for Irregular Problems. J. of Supercomputing 10 (2) (1996) 119–140.Google Scholar
- 6.Woo, S.C., Ohara, M., Torrie, E., Singh, J.-P., Gupta, A.: The SPLASH2 Programs: Characterization and Methodological Considerations. ACM/IEEE Int’l. Symp. Computer Architecture (ISCA’95) (Jun. 1995) 24–26.Google Scholar