Implementation and performance of a domain decomposition algorithm in Sisal
- Lawrence Livermore National Lab., CA (United States)
- California Univ., Livermore, CA (United States)
- State Univ. of New York, Stony Brook, NY (United States)
Sisal is a general-purpose functional language that hides the complexity of parallel processing, expedites parallel program development, and guarantees determinacy. Parallelism and management of concurrent tasks are realized automatically by the compiler and runtime system. Spatial domain decomposition is a widely-used method that focuses computational resources on the most active, or important, areas of a domain. Many complex programming issues are introduced in paralleling this method including: dynamic spatial refinement, dynamic grid partitioning and fusion, task distribution, data distribution, and load balancing. In this paper, we describe a spatial domain decomposition algorithm programmed in Sisal. We explain the compilation process, and present the execution performance of the resultant code on two different multiprocessor systems: a multiprocessor vector supercomputer, and cache-coherent scalar multiprocessor.
- Research Organization:
- Lawrence Livermore National Lab., CA (United States)
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 10191698
- Report Number(s):
- UCRL-JC--115029; CONF-940136--3; ON: DE94002012
- Country of Publication:
- United States
- Language:
- English
Similar Records
Proceedings: Sisal `93
An implementation of SISAL for distributed-memory architectures