Automatic Counterflow Pipeline SynthesisReport
The Counterflow Pipeline (CFP) organization may be a good target for synthesis of application-specific microprocessors because it has a regular and simple structure. This paper describes early work using CFP's to improve overall application performance by tailoring a CFP to the kernel loop of an application. A CFP is customized for an application using the kernel loop's data dependency graph to determine processor functionality and interconnection. Our technique builds the design space for a given data dependency graph and explores the space to find the design with the best performance. Preliminary results indicate that speed-up for several small graphs range from 1.3 to 2.0 and that our design space traversal heuristics find designs that are within 10% of optimal.
All rights reserved (no additional license for public reuse)
Childers, Bruce, and Jack Davidson. "Automatic Counterflow Pipeline Synthesis." University of Virginia Dept. of Computer Science Tech Report (1998).
University of Virginia, Department of Computer Science