Programming Languages Lunch - Jim Browne, Leonardo Fialho, and Ashay Rane, UTCS, "Unification and Automation of Compile Time and Runtime Performance Optimization"

Contact Name: 
John Thywissen
GDC 6.302
Nov 25, 2013 12:00pm - 1:00pm

Talk Audience: UTCS Faculty, Grads, Undergrads, Other Interested Parties

Talk Abstract: Even the best modern compilers cannot, using only the information available from static analyses, always generate optimal executable code for today's highly complex multicore and manycore chips and heterogeneous compute nodes. Performance optimization based on runtime information is thus often required to make best use of complex system architectures. The PerfExpert project, which includes the MACPO (Memory Access Centric Performance Opimization) tool, brings two significant innovations to performance optimization: automation of end-to-end performance optimization and integration of metrics specifying data structure memory access patterns with traditional code segment specific resource use metrics in performance optimization. While compile time and runtime performance optimization have traditionally been separate research communities, the PerfExpert project targets developing and implementing a unification of these two research disciplines. Intel has seen the benefit of this approach and has provided additional funding for the project. This talk will focus on what has been accomplished by the PerfExpert project including MACPO and the many research opportunities created by unification of runtime and compile time performance optimization.

Speaker Bio: Jim Browne is Professor Emeritus of Computer Science and Physics and at The University of Texas at Austin and Chief Techology Officer for the Stampede computer at the Texas Advanced Computing Center. He was, from 1965 through 1968, Professor of Computer Science at Queens University in Belfast and directed the Computer Laboratory.  Browne rejoined The University of Texas in 1968 as Professor of Physics and Computer Science. He served as Department Chair for Computer Science for three terms and became Emeritus in 2008.

Leonardo Fialho was doing research on fault tolerance for MPI applications at the University Autonoma of Barcelona, Spain, where he obtained an MSc and PhD degrees. His first postdoctoral position was at Dublin City University (DCU), Ireland, doing research on power efficiency for telecommunication networks. Currently, Leonardo is an associate researcher at the Texas Advanced Computing Center at the University of Texas at Austin doing research on performance of parallel applications where he is currently the chief designer and implementer of PerfExpert.

Ashay Rane is a PhD student in the Computer Science department at the University of Texas at Austin. Prior to joining the PhD program in the Fall of 2012, he researched, developed and maintained PerfExpert from December 2010 until August 2012. He graduated with a Master's degree in Computer Science from Arizona State University. His areas of interest include compilers, computer architecture and performance tuning.