Title/Author/Abstract

• Title:
  Parallel Object-Oriented Framework Optimization

• Author:
  Dan Quinlan

• Full Paper(.ps version)

• Abstract:
  The development of optimizations for parallel object-oriented frameworks is complicated by the inability of the compiler to know about the framework's abstractions (objects). The compile time program analysis can be simplified though the use of the semantics of a framework's abstractions, but such an approach requires the use of a preprocessor mechanism ahead of the language compiler or a specialized compiler more directly. We will present current work on optimizations using a preprocessor approach. Within our approach specialized grammars (automatically generated from the class declarations) are used to recognize user defined objects within scientific applications. For objects with parallel semantics parallel optimizations are possible. This talk will present some of the current work being done to optimize the performance of parallel object-oriented frameworks used for scientific computing.

Home : Cpc2001