Thursday, January 25, 2001
4:00 PM in 338 Eberly, Refreshments at 3:30 PM
Dr. Charles Lefurgy
IBM Austin Research Lab
Systems Software Group
ABSTRACT
- My research examines methods to reduce program code size using data compression techniques. The motivation for this research is to reduce the die cost of systems-on-a-chip by enabling them to execute programs in a small memory footprint. After compilation, binary program images are compressed. At run-time, the programs are incrementally decompressed and executed. I have simulated doing the decompression with both hardware and software and developed optimizations for each.
- In this talk I will present my work on executing compressed programs using software decompression (with minimal hardware assistance). I will show results for two software optimizations (hybrid programs and memoization) that improve the execution time of compressed programs. Finally, I will demonstrate that software decompression can have low overhead. In fact, loop-oriented (media) programs have performance levels similar to native code.
BIOGRAPHY
- Charles Lefurgy is a Research Staff Member in the Systems Software group of the IBM Austin Research Lab. He joined IBM in September 2000 and studies software strategies for low-energy computers. His other research interests include microarchitecture, memory systems and compilers. Previously, he completed a Ph.D. in computer engineering at the University of Michigan. His dissertation work focused on compressed memory systems for embedded computing.