Monday, March 17

11:00 AM-1:00 PM
Nicollet D3

Generating Efficient Parallel Scientific Code From High-Level Descriptions

Programmer productivity is a major concern, in particular in the creation of codes of high performance for scalable architectures. High-level descriptions offer high productivity by having high expressitivity. But such descriptions are useful in high performance computation only if efficient code can be generated.

Domain-specific formal languages and visual programming languages are but two of the approaches to achieving productivity. Domain-specific languages, for example, allow the programmer to describe a program in a notation close to or identical to the one used in the discipline, but create a serious challenge for compiler technology.

The speakers in this minisymposium will present some current attempts to compile efficient parallel code for scientific applications and report on early accomplishments.

Organizer: S. Lennart Johnsson
University of Houston and Harvard University

11:00 A Kronecker Compiler for Fast Transform Algorithms
Nikos P. Pitsianis, University of Houston
11:30 Automatic Parallelization of Sparse Matrix Applications
Vladimir Kotlyar, Keshav Pingali, and Paul Stodghill, Cornell University
12:00 Efficient Data Structures for Sparse Iterative Methods
Mark T. Jones, University of Tennessee, Knoxville and Paul E. Plassmann, Argonne National Laboratory
12:30 Optimization of a Class of Multi-Dimensional Integrals on Parallel Machines
C. Lam, P. Sadayappan, and R. Wenger, Ohio State University, Columbus

PP97 Homepage | Program Updates|
Registration | Hotel Information | Transportation | Program Overview | Program-at-a-Glance

MMD, 1/24/97