[PAST EVENT] Computer Science Colloquium: Peng Wu

October 4, 2013
McGlothlin-Street Hall, Room 020
251 Jamestown Rd
Williamsburg, VA 23185Map this location
Evolution of a compiler expert in an industry research lab: SIMD, Scripting Languages, Data Analytics, and more

What is it like to do research in an industry research lab? In this talk, I will reflect on the evolution of my decade-long career as a compiler expert in IBM research. As it is sometimes revealing to examine a line of research over an extended period of time, I will focus on a particular research of mine on SIMD exploitation that has spanned well over a decade (with big gaps in between). We tell the story from the eyes of a compiler expert who was heavily involved in the programming model design of the CELL SIMD engine from 2001 to 2006. When returning to SIMD to design the next generation POWER processor in 2012, she found the world of SIMD has completely changed. The technical part of the talk will cover the following topics:

What is the concept of SIMT (Single-Instruction-Multiple-Thread)?

What is the software and hardware landscape of SIMD today and where will it be in 5 years?

The surprising role of a compiler researcher in a hardware project.

Apart from SIMD, I will present some of the other important problems we are solving today in the field of system compiler research. And I will conclude with my thoughts on the broader role of compiler researchers in the industry and the important lessons I learned.

Peng Wu joined IBM T.J. Watson research center as a research staff member in 2001 after receiving her Ph.D. in computer science from University of Illinois, Urbana-Champaign. Her work has expanded across all layers of the system stack including architecture, compiler/runtime, programming interface, and whole-system optimization. She made major contribution to IBM's product C/C++/Fortran compilers and Java VM/JIT. She has also worked extensively with hardware designers including POWER, the CELL BE, and BlueGene L/P. Her current research focuses on scripting languages runtime/JIT such as R, V8, and Python; large-scale graph runtime; SIMD and parallelism; and workload-specific optimizations. She held more than 20 patents, and has co-authored more than 30 papers. Her work on hardware transactional memory won a best paper award in PACT 2012. She is an adjunct professor in the Department of Computer Science at University of Illinois, Urbana-Champaign since 2012.

[[cs|xshen,Xipeng Shen]]