COLLOQUIUM Department of Computer Science and Engineering University of South Carolina Synthesizing DSP Applications onto Reconfigurable Computing Systems Wenrui Gong Department of Electrical and Computer Engineering University of California, Santa Barbara Date: March 16, 2006 Time: 1100-1200 Place: Swearingen 1A03 (Faculty Lounge) Abstract Modern reconfigurable computing systems feature powerful hybrid architectures with large reconfigurable logic arrays, distributed memory hierarchies, dedicated DSP devices and embedded microprocessor cores. These systems are widely used in telecommunications, multimedia, digital entertainment, sensing, etc. Reconfigurable systems can provide more flexibility, lower design cost and shorten time-to-market. This talk focuses on the architectural-level design methodology that we used to map DSP applications to FPGA-based reconfigurable systems. We utilized loop transformation and storage assignments to create coarse-grained parallelism, and developed the ACO algorithm, an evolutionary algorithm, to exploit the design space and generate the fastest design for each thread. Several memory optimization techniques, such as pre-fetching, scalar replacement, buffer insertion, were investigated. Experiments showed that these techniques could greatly benefit generated designs and help designers achieve performance goals. Wenrui Gong is a Ph.D. candidate in Electrical and Computer Engineering at the University of California, Santa Barbara, where he received his M.S. degree in Electrical and Computer Engineering in December 2002. In July 1999, he received a B.Eng. degree in Computer Science from Sichuan University, China. Mr. Gong’s research interests include architectural-level synthesis of integrated circuits and systems, parallelizing compilation techniques, novel reconfigurable architectures, and combinatorial optimization algorithms and their application in design automation. In the summers of 2004 and 2005, he interned at Mentor Graphics and worked on the Catapult C Synthesis tool.