Highly Concurrent Scalar Processing.
Abstract
High speed scalar processing is an essential characteristic of high performance general purpose computer systems. Efficient concurrent execution of scalar code is difficult due to data dependencies and conditional branches. This thesis proposes a code scheduling heuristic call the decision tree scheduling (DTS) technique for general scalar code, and an optimal code scheduling algorithm called the simple loop scheduling (SLS) algorithm for a restricted class of innermost loops. Also proposed is a highly concurrent machine architecture that takes advantage of these scheduling techniques. Keywords: parallel processing; throughput; instructions; multiprocessors.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1986
- Accession Number
- ADA163195
Entities
People
- Peter Y. Hsu
Organizations
- University of Illinois Urbana–Champaign