This shows you the differences between two versions of the page.

Link to this comparison view

seminars:seminar_2_16_16 [2017/09/20 22:02] (current)
Line 1: Line 1:
 +====== Architectural Specialization for Inter-Iteration Loop Dependence Patterns ======
 +Tuesday February 16, 2016\\
 +Location: CIC Panther Hollow Room\\
 +Time: 4:30PM\\
 +**[[http://​www.csl.cornell.edu/​~cbatten/​|Christopher Batten (Cornell University)]]**\\
 +Hardware specialization is an increasingly common technique to enable
 +improved performance and energy efficiency in spite of the diminished
 +benefits of technology scaling. Exploring hardware specialization
 +requires a vertically integrated research approach spanning
 +applications,​ compilers, run-times, instruction set design,
 +microarchitectures,​ and VLSI implementation.
 +In the first part of the talk, I will describe our work on a new
 +architectural design pattern called explicit loop specialization
 +(XLOOPS) based on the idea of elegantly encoding inter-iteration loop
 +dependence patterns in the instruction set. The XLOOPS
 +hardware/​software abstraction requires only lightweight changes to a
 +general-purpose compiler to generate XLOOPS binaries and enables
 +executing these binaries on: (1) traditional microarchitectures with
 +minimal performance impact, (2) specialized microarchitectures to
 +improve performance and/or energy efficiency, and (3) adaptive
 +microarchitectures that can seamlessly migrate loops between
 +traditional and specialized execution to dynamically trade-off
 +performance vs. energy efficiency. Our initial results show promising
 +performance improvements compared to simple in-order processors and
 +energy efficiency improvements compared to complex out-of-order
 +In the second part of the talk, I will describe PyMTL and Pydgin, two
 +new Python-based frameworks designed to improve the productivity of
 +vertically integrated computer architecture research. PyMTL leverages
 +the Python programming language to create a highly productive
 +domain-specific embedded language for concurrent-structural modeling
 +and hardware design. The PyMTL framework encourages a philosophy of
 +"​modeling towards layout"​ in which a microarchitecture is
 +incrementally refined from a high-level functional-level model, to a
 +timing-approximate cycle-level model, to a bit-accurate RTL
 +implementation. Pydgin is a framework for rapidly developing very fast
 +instruction-set simulators (ISSs) from a Python-based architecture
 +description language. Both frameworks were critical in exploring the
 +XLOOPS architectural design pattern, and we are continuing to leverage
 +these frameworks in our research and teaching.
 +Christopher Batten is an Assistant Professor in the School of
 +Electrical and Computer Engineering at Cornell University, where he
 +leads a research group focusing on energy-efficient parallel computer
 +architecture for both high-performance and embedded applications. His
 +work has been recognized with several awards including an AFOSR Young
 +Investigator Program award (2015), Intel Early Career Faculty Honor
 +Program award (2013), an NSF CAREER award (2012), a DARPA Young
 +Faculty Award (2012), and an IEEE Micro Top Picks selection (2004).
 +His teaching has been recognized with a Michael Tien '72 Excellence in
 +Teaching Award (2013) and a James M. and Marsha D. McCormick Award for
 +Outstanding Advising of First-Year Engineering Students (2013). Prior
 +to his appointment at Cornell, Batten received his Ph.D. in electrical
 +engineering and computer science from the Massachusetts Institute of
 +Technology in 2010. He received an M.Phil. in engineering as a
 +Churchill Scholar at the University of Cambridge in 2000, and received
 +a B.S. in electrical engineering as a Jefferson Scholar at the
 +University of Virginia in 1999.
 +**[[seminars| Back to the seminar page]]**