Compiling for Locality in Advanced Scientific Computations

University Of Maryland, College Park, College Park MD

Investigators

Abstract

With modern processor architectures, programs can achieve good performance only if they possess sufficient data locality to exploit on-chip caches. This research focuses on developing and evaluating software support for improving locality for advanced scientific applications for both sequential and parallel machines. The basic premise is that both compile-time analyses and sophisticated run-time systems are necessary. Run-time systems are needed because many programs are not analyzable statically. Compiler support is crucial both for inserting interfaces to the run-time system and for directly applying program transformations where possible. This proposal investigates locality optimizations needed for three features found in advanced scientific applications (3D arrays, irregular accesses, and pointers). It focuses on extending locality optimizations to handle cache conflicts between multiple data, deep memory hierarchies (multi-level caches and TLBs), hardware and software prefetching, nonlinear memory layouts, parallel and cluster architectures, and memory performance tools. Locality optimizations will be applied to representative programs and experimentally evaluated on advanced computer systems. The results of this research should help improve the performance of computationally intensive scientific applications. Because of trends in computer architectures, lessons learned are also likely to be useful for application domains such as image processing and high-performance databases.