home · mobile · calendar · colloquia · 1999-2000 · 

Colloquium - Connors

Eliminating Dynamic Computation Redundancy Using a Compiler-Directed Approach
University of Illinois at Urbana-Champaign

One of the major challenges to increasing processor performance is overcoming the fundamental dataflow limitation imposed by data dependences. By reusing previous computation results, this inherent dataflow limit can be exceeded for sequences of operations that are otherwise redundantly executed. Several empirical studies indicate that significant amounts of redundancy still exist in optimized programs. One architectural paradigm presented to eliminate dynamic computation redundancy is the Compiler-Directed Computation Reuse (CCR) approach. The CCR approach integrates compiler and architecture techniques in a coordinated framework to eliminate large amounts of redundant computation. The framework allows the compiler to identify code regions whose computation results can be reused during dynamic execution. During run time, the CCR architectural components record the execution results of the regions for potential reuse, eliminating the use of processor resources in future executions. This talk will describe work which enhances the CCR approach by developing an effective run-time monitoring and management system to dynamically control the deployment of statically-selected computation regions. The Computation Management system increases the effectiveness of the approach by systematically adjusting the utilization of the reuse caching resources to the dynamic program behavior. An analysis of the enhanced CCR approach, showing the approach's ability to eliminate dynamic redundancy and improve performance, will be presented.

Hosted by Dirk Grunwald.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)