Advanced Compilation Support for Superscalar and VLIW Processors
Superscalar and VLIW processors van potentially provide large performance improvements over their scalar predecessors by concurrently executing independent instructions. However, these promised performance improvements often cannot be realized with conventional complier optimization and scheduling techniques. In the first half of this advanced seminar, we will discuss new features in the IMPACT complier to uncover instruction level parallelism (ILP) for the current commercial superscalar microprocessors. The features include a detailed machine description database for use by optimization and code scheduling, identifying safe instructions for speculative execution, code optimization for conditional moves, superblock code optimization and scheduling, hyperblock code optimization and scheduling with reverse if-conversion, software pipelining, and deriving high level memory dependence information for use in optimization and scheduling.
A. Ferscha received the Mag. degree in 1984, and a Ph.D. in business Informatica in 1990, both from the University of Vienna, Austria. In 1986, he joined the department of Applied Computer Science at the University of Vienna, where he is presently an Assistant Professor. He has been a lecturer in parallel processing and performance evaluation since 1988. His current research interests include performance modeling of parallel systems, communication structures in parallel programs, computer aided engineering or parallel software, distributed simulation strategies and parallel neural network simulation.
Introduction To Protein Structure PredictionProtein structure prediction is the premier computational problem confronting molecular biology today. This sessionr, intended for the non-biologist, will survey the different faces of the problem and a number of approaches to it. We begin with an object-oriented introduction to the behavior of amino acid, peptides, proteins and water, avoiding deep physics or chemistry. This motivates what the problem is and why it is hard, and allows us to understand the multitude of approaches as solutions to more trackable sub-problems. Topics will include direct and inverse folding methods, homology-derived techniques, tertiary topology (domain) prediction, secondary structure prediction, lattice folding models, Monte Carto and modeling, amino acids environments, systhetic energy potentials, and other topics of interest.
Richard Lathrop conducts research in computational biology at the Artificial Intelligence Laboratory of MIT in consortium with the BioMolecular Engineering Research Center of Boston University. He received his Ph. D in Computer Science from MIT in 1990. He is on the organizing committees of a variety of conference and workshops on computer science and molecular biology, and is a founding scientist of Arris Pharmaceutical Corporation. His research publications include several book chapters and numerous journal and conference articles in protein structure/function analysis, biological sequence analysis, massively parallel machine learning, robotics, and VLSL design.
Parallel Programming Tools-Status, Evaluation, & ComparisonThis seminar addresses the complexity of developing parallel programs, describes the characteristics of scientific applacations and their developers, and the computing environment in a typical high performance couputing center. User requirements for tools that support application portability and present the difficulties to satisfy them are discussed. Tools in the following categories are compared and evaluated: 1)converting sequential programs to parallel programs, 2) developing portable new parallel programs, 3) debugging and performance tuning, 4)partitioning and mapping, and 5) managing network of resources. The future directions of the tools in each category will be discussed.
Doreen Y. Cheng received her Ph. D. in Electrical Engineering from Stanford University in 1988. She is currently at NAS/NASA Ames Research Center and leading a team to develop tools for writing, debugging and optimizing parallel programs. Her NASA report, A Survey of Parallel Programming Languages and Tools has been requested by hundreds of organizations worldwide and is highly regarded. Together with colleagues, she is defining protocols that facilitate debugger portability and developing a prototype for a portable parallel/distributed debugger.
Performance Analysis Of Parallel SystemsPerformance analysis has become a critical issue in parallel system/program development. This seminar presents fundamental performance analysis and evaluation principles for parallel computing. The topics include: 1) estimates and bounds on the execution time of a parallel application, 2) measurements (monitoring, benchmarking) performed on real systems, and 3) performance modeling, i.e. trying to characterize a real or hypothetical system by a set of mathematical terms (analytical models), or by a set of states and events (simulation models). The corresponding evaluation technique are described and their impact on the parallel program development processes is investigated. Tools supporting the various kinds of analysis are presented. Finally, a development process for parallel programs that launches performance engineering in the early design phase is proposed.
A. Ferscha received the Mag. degree in 1984, and a Ph. D in business information in 1990, both from the University of Vienna, Austria. In 1986, he joined the department of Applied Computer Science at the University of Vienna, where he is presently an Assistant Professor. He has been a lecturer in parallel processing and performance evaluation since 1988. His current research interests include performance modeling of parallel systems, communication structures in parallel programs, computer aided engineering or parallel software, distributed simulation strategies and parallel neural network simulation.
HICSS-28 HOMEPAGE
HICSS HOMEPAGE