Upcoming Seminars
Wednesday, October 8, 2008
Silicon Nanophotonics:
A Platform for High-Performance On-Chip Optical Interconnects
Dr. William Green
IBM Thomas J. Watson Research Center, Yorktown Heights, NY
Monday, October 13, 2008
Current and Next-Generation Passive Optical Networks
Dr. Pat Iannone
AT&T Labs - Research
Past Seminars
October 6, 2008
The World of Quantum Information
Professor Marianna Safronova
Department of Physics and Astronomy, University of Delaware
October 3, 2008
Contrast-Enhanced Subharmonic Imaging
- A New Diagnostic Tool
Flemming Forsberg, Ph.D.
Department of Radiology, Thomas Jefferson University
September 29, 2008
A Game-Theoretic Analysis of Decode-and-Forward User Cooperation
Professor Shalinee Kishore
Lehigh University
September 24, 2008
How Does DARPA Work?
Michael Haney, Professor
Electrical & Computer Engineering, University of Delaware
September 15, 2008
Robust Sampling and Reconstruction Mehtods for Sparse Signals in the Presence of Impulsive Noise
Rafael Carrillo
University of Delaware
September 12, 2008
Investigating Solute Transport and Cell-to-Cell Signaling in bone
Liyun Wang, Ph.D.
Mechanical Engineering, University of Delaware
July 24, 2008
Iterative Compilation by Exploration of Kernel Decomposition
William Jalby, Professor
University of Versailles Saint Quentin
May 19, 2008
Recent Advances in Computation Photography
Jingyi Yu, Assistant Professor
Computer and Information Science Department University of Delaware
May 7, 2008
Gauss' Law: What does it NOT say?
H. Brian Sequeira
Johns Hopkins University, Applied Physics Laboratory
April 30, 2008
Circuits with Light at the Nanoscale
Nader Engheta
H. Nedwill Ramsey Professor, University of Pennsylvania
Department of Electrical and Systems Engineering, Philadelphia, Pennsylvania
April 15, 2008
Vacuum Technology Seminar
Varian Inc. Vacuum Technologies
March 7, 2008
Microfluidic Cell Arrays for High Throughput Signal Pathway Profiling and Drug Screening
Professor Sihong Wang
Department of Biomedical Engineering, City University of New York
February 15,
Flow Control on the Micro-Scale
Dr. Benjamin Shapiro, Associate Professor
University of Maryland, Department of Aerospace Engineering
December 5, 2007
High Power Diode Lasers and Applications
Dr. Wei Gao, President and CTO
Axcel Photonics
November 30, 2007
Fascinating Rhythms:
Reverse Engineering Cortical Function from Changes in Brain Oscillations in Neurological and Psychiatric Disease
Leif Finkel, MD, PhD
University of Pennsylvania, Department of BioEngineering
November 26, 2007
Molecular Level Modeling and Dynamic Analysis of Biochemically Coupled Multicellular Systems
Michael Henson
University of Massachusetts Amherst
November 21, 2007
Nanostructures for Potential Signal Transduction Constructed via Molecular Self-Assembly
Dr. Darrin Pochen
University of Delaware, Materials Science and Engineering
November 19, 2007
Optimal Precoding for Multiple-Input Multiple-Output Gaussian Channels with Arbitrary Inputs
Dr. Fernando Perez-Cruz
Universidad Carlos III & Princeton University (joint work with Miguel Rodrigues and Sergio Verdu)
November 12, 2007
Random Control Bounds for Block Coded Transmission over Fading MIMO
Multiple Access Channel
November 7, 2007
Metal/Semiconductor Nanocomposites
Professor Joshua Zide
University of Delaware
November 2, 2007
Engineering Cancer Therapies:
Mathematical Modeling of Tumor Metabolism
and Therapeutic Efficacy
Professor Neil Forbes
University of Massachusetts Amherst
October 31, 2007
Finite Difference Delay
Dr. Xiaobo Wang
University of Delaware
October 19, 2007
Scheduling of Optimal Medication Strategies for Early HIV Infection
Professor Antonios Armaou
Pennsylvania State University
October 17, 2007
Is There Life on the Moon?
Professor Brian Sequeira
Johns Hopkins APL
April 30, 2007
Distributed Processing over Adaptive Networks
Professor Ali H. Sayed
University California, Los Angeles
April 9, 2007
Thoughts on Innovation
Ray Sokola,
Chief Technology Officer Motorola Inc.
Monday, March 12, 2007
Packet Switching Comes of Age: From Research to Commercial Development
Dr. W. David Sincoskie
March 5, 2007
Wireless Sensor Networks
Professor Edward Coyle,
Purdue University
Wednesday February 14, 2007
Confessions of an Internet Timekeeper
Dr. David L. Mills
The University of Delaware, Department of Electrical and Computer Engineering
ITERATIVE COMPILATION BY EXPLORATION OF KERNEL DECOMPOSITION
William Jalby, Professor
University of Versailles Saint Quentin
10:30 AM Evans Hall, Room 204
Abstract:
The increasing complexity of hardware features for recent processors makes high performance code generation very challenging. General purpose compilers, with no knowledge of the application context and no accurate performance models, seem inappropriate for this task. On the other hand, combining application-dependent optimizations on the source code and exploration of optimization parameters as it is achieved with ATLAS, has been shown as a promising path to achieve high performance. Yet, hand-tuned codes such as in the MKL library still outperforms ATLAS with an important speed-up and some effort has to be done in order to bridge the gap between performance obtained by automatic tools (e.g ATLAS) and manual optimizations.
We propose a new iterative compilation approach for the generation of high performance codes relying on the use of state of the art compilers. At the opposite of ATLAS, this approach is not application-dependant (i.e limited to one type of algorithm) but can be applied to fairly generic loop structures. In a classical manner, the memory optimization phase is decoupled from the computation optimization phase. First the loop is blocked to obtain computational primitives fitting in the cache. The second step aims at finding automatically all possible decompositions of the code into simpler code fragments (typically 1 or 2 dimensiuonnal loops) called kernels. With datasets that fit into the cache and simplified memory accesses, these kernels are simpler to optimize, either with the compiler, at source level, or with a dedicated code generator. The best decomposition is then found by a model-guided approach, performing on the source code the required memory optimizations.
Exploration of optimization sequences and their parameters is achieved with a meta-compilation language, X language. The first results on linear algebra codes and for two fairly different architectures (Itanium II and Pentium 4) show that the performance obtained reduce the gap with those of highly optimized hand-tuned codes.
BIOGRAPHY
William JALBY was appointed Associate Professor at University of Rennes in 1987, then promoted Full Professor of Computer Science in 1991 and moved in 1992 to University of Versailles. His areas of research are:
performance evaluation, code optimization, memory hierarchies and embedded processing. From 1987 to 1992, W. Jalby has been working closely with CSRD (CEDAR project, University of Illinois). More recently, he is collaborating with CEA DAM (French equivalent of Los Alamos) on performance evaluation and with BULL on code optimization for Itanium based SMP. He is the head of a joined Laboratory (LRC ITACA) between CEA DAM and University of Versailles, specialized in code optimization.