Architecture and Code Optimization Techniques to Create Fast and Effective Data Compressors
Tuesday October 17, 2006
Hamerschlag Hall 1112
This talk illustrates how to exploit ideas from microarchitectural and compiler
research to achieve breakthroughs in lossless floating-point and trace compression,
which in turn benefit architecture and code optimization researchers. We start with a
brief overview of lossless data compression. Then we show how hardware prediction
schemes can be used to design trace compression algorithms that typically outperform
other approaches. Next, we present the compiler optimization techniques we employed
to automatically generate such compressors and to tailor them to user-provided trace
formats. Finally, we discuss the necessary modifications to turn our algorithm into
one of the fastest and most effective lossless floating-point compression algorithms.
Martin Burtscher 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 an assistant
professor in the School of Electrical and Computer Engineering at Cornell University,
where he leads the High-Performance Microprocessor Systems Group. His present 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 is a senior member of the IEEE and its Computer Society and a member of the ACM.