Systems Software for Irregular Parallel Applications
Abstract
The long-term objective of this project was to improve programmability of parallel machines for irregular applications with unpredictable computational costs, pointer-based data structures, dynamically allocated data structures, Or asynchronous communications. Example applications include sparse matrix algorithms, adaptive mesh refinement algorithms, and symbolic algorithms. The trend in recent years toward deeper memory hierarchies, including several levels of cache, DRAM, network, and disk, has meant that the more irregular applications have not yet benefited from increasing processor speed as much as more regular applications. Our goal was to provide programmers with high level tools for writing their high performance applications, and our specific tasks addressed three aspects of this problem: application understanding, library development, and compiler optimizations.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 24, 2001
- Accession Number
- ADA391106
Entities
People
- Katherine Yelick
Organizations
- University of California, Berkeley