Using Architecture and Code Optimization Techniques to Create Fast and Effective Data Compressors

December 4, 2006
Martin Burtscher | Cornell University

This talk illustrates how to exploit computer architecture and compiler optimization ideas to achieve breakthroughs in lossless program-trace and floating-point compression, which in turn benefit architecture and compiler researchers. We start with an overview of value prediction hardware. Then we show how these prediction concepts can be used to design trace compression algorithms that typically outperform other approaches in both speed and compression ratio. Next, we present the compiler optimization techniques we employed to automatically synthesize such compressors and tailor them to user-specified trace formats. Finally, we discuss the necessary modifications to convert our algorithm into a fast and effective lossless floating-point compression algorithm.

Speaker Details

Martin Burtscher is an assistant professor in the School of Electrical and Computer Engineering at Cornell University, where he leads the High-Performance Microprocessor Systems Group. His research focuses on speeding up individual program threads using multiple processor cores, designing computer systems that autonomously adapt to the current workload, improving the performance of sequential and parallel programs, and creating fast and effective data compression algorithms. He received the combined BS/MS degree in computer science from the Swiss Federal Institute of Technology (ETH) Zurich in 1996 and the PhD degree in computer science from the University of Colorado at Boulder in 2000. He is a senior member of the IEEE and the ACM.