Many important computational problems posed by scientists and engineers require harnessing the power of parallel computers to provide timely solutions. Today, parallel systems are most commonly built from ensembles of commodity microprocessors. Such systems are notoriously difficult to program effectively. Key stumbling blocks to using parallel computers productively include primitive programming systems and the lack of capable program development tools.

This talk will describe research on compiler technology to support more productive programming models for microprocessor-based parallel systems, as well as compiler-based tools for measuring and modeling the interplay between applications and hardware to gain insight into how to construct better applications and computer systems. First, I will describe data parallel compilation techniques to support abstract models for high-performance parallel programming. Second, I will describe a compiler-based toolkit for semi-automatic synthesis of scalable performance models, which have applications ranging from scheduling computations on heterogeneous grids to workload-based design of future computer systems. The talk will conclude with an assessment of progress to date and challenges for the future.

Dr. Mellor-Crummey is a faculty candidate. Brief biography: John Mellor-Crummey is a Senior Faculty Fellow in Computer Science and Electrical & Computer Engineering at Rice University. He received his PhD and MS degrees from the University of Rochester in 1989 and 1986, respectively, and his Bachelors degree from Princeton University in 1984.

Monday April 19, 2004

Reception 2:30 p.m. in DH 3092.
Presentation follows at 3:00 p.m. in DH 1070.