Select a calendar:
Filter April Events by Event Type:
SUNMONTUEWEDTHUFRISAT
Events for April 17, 2014
-
EE-Systems Seminar
Thu, Apr 17, 2014 @ 10:30 AM - 12:00 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Wonsun Ahn , Postdoctoral Research Scientist, University of Illinois at Urbana-Champaign
Talk Title: High-Performance JIT-Compiled Frameworks: Hardware/Compiler Co-Optimization
Abstract: JIT-compiled frameworks are gaining increasing use for their cross-platform portability, performance portability, and runtime adaptability. In particular, scripting languages such as JavaScript, Python, and R are gaining wide acceptance. In these emerging frameworks, there is great opportunity for performance improvement through hardware/compiler co-optimization. In this talk, I present a few novel techniques that I have developed to improve performance.
First, I show how the compiler can use Hardware Transactional Memory (HTM) support to enforce high-performance Sequential Consistency (SC) for programmability and security. The idea is to wrap large sections of code inside a transaction, and then optimize the code inside each transaction without concern for memory-consistency-model restrictions. The optimizations speculate that any violation of the memory model will not be seen by other threads; otherwise, the transaction is aborted. Using this approach, the compiler can even outperform current compilers by a significant margin by allowing optimization across synchronization boundaries.
Next, I also show how the compiler can use the same HTM support to perform alias speculation. The approach consists of performing optimizations assuming the alias relationships that are true most of the time, and using the hardware to detect when such relationships are found not to hold through runtime checks. If the assumptions are correct, the code experiences good speedups; otherwise, the transaction is aborted.
Lastly, I show a compiler enhancement for JavaScript. A key feature of scripting languages that gives them their flexibility is dynamic typing. However, the absence of declared types makes it very challenging for the compiler to generate efficient code. Advanced compilers cope with it by introducing type systems of their own behind the scenes, and maintaining the type of each object at runtime as metadata. In this work, I focus on the Google Chrome V8 JavaScript compiler, and show that its type system is too brittle. While it works well for applications that display static behavior, it causes type specialization to fail in real website code. I go on to modify V8's type system to match the more dynamic behavior of real websites, and show significant savings in execution time, energy, and memory consumption.
Overall, these three approaches allow JIT compilers to achieve high performance while still maintaining programmability and security.
Biography: Wonsun Ahn is a Postdoctoral Research Scientist at the University of Illinois at Urbana-Champaign. His research interests are parallel computer architecture and compilation systems. He is currently the co-PI of an NSF grant on improving the performance of scripting languages. He received a PhD in Computer Science from the same university in 2012. His PhD work was recognized by an IEEE Micro's Top Picks award publication. He has (co-)authored 12 journal and conference papers that have appeared in top compiler and architecture venues, and has two industry patents. He has served in the program and organizational committees of conferences, and is a member of the Samsung Frontier Membership.
Host: Michel Dubois
More Information: print_Ahn.pdf
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248
Audiences: Everyone Is Invited
Contact: Estela Lopez
-
Andrew J. Viterbi Distinguished Lecture in Communication
Thu, Apr 17, 2014 @ 04:00 PM - 05:00 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Professor Abbas El Gamal, Stanford University
Talk Title: Common Information
Series: Distinguished Lecturer Series
Abstract: Entropy, introduced by Shannon in 1948, arises naturally as a universal measure of information in single-source compression, randomness extraction, and random number generation. In distributed systems, such as communication networks, multiprocessors, distributed storage, and sensor networks, there are multiple correlated sources to be processed jointly. The information that is common between these sources can be utilized, for example, to reduce the amount of communication needed for compression, computing, simulation, and secret key generation. My talk will focus on the question of how such common information should be measured.
While our understanding of common information is far from complete, I will aim to demonstrate the richness of this question through the lens of network information theory. I will show that, depending on the distributed information processing task considered, there can be several well-motivated measures of common information. Along the way, I will present some of the key models, ideas, and tools of information theory, which invite further investigation into this intriguing subject.
Some parts of this talk are based on recent joint work with Gowtham Kumar and Cheuk Ting Li and on discussions with Young-Han Kim.
Biography: Abbas El Gamal is the Hitachi America Professor in the School of Engineering and Chair of the Department of Electrical Engineering at Stanford University. He received his Ph.D. degree in electrical engineering from Stanford University in 1978. He was an Assistant Professor in the Department of Electrical Engineering at the University of Southern California (USC) from 1978 to 1980. His research interests and contributions have spanned the areas of information theory, wireless networks, CMOS imaging sensors and systems, and integrated circuit design and design automation. He has authored or coauthored over 200 papers and 30 patents in these areas. He is coauthor of the book Network Information Theory (Cambridge Press 2011). He has won several honors and awards, including the 2012 Claude E. Shannon Award, and the 2004 Infocom best paper award. He is a member of the National Academy of Engineering and a Fellow of the IEEE. He has been active in several IEEE societies, including serving on the Board on Governors of the IT society where he is currently its President. He cofounded and/or served in various leadership roles at several semiconductor, EDA, and biotechnology companies.
Host: Professor Sandeep Gupta
Location: Seeley Wintersmith Mudd Memorial Hall (of Philosophy) (MHP) - 101
Audiences: Everyone Is Invited
Contact: Mayumi Thrasher
