Optimizing chip multiprocessor work distribution using dynamic compilation

Abstract

How can sequential applications benefit from the ubiquitous next generation of chip multiprocessors (CMP)? Part of the answer may be a dynamic execution environment that automatically parallelizes programs and adaptively tunes the work distribution. Experiments using the Jamaica CMP show how a runtime environment is capable of parallelizing standard benchmarks and achieving performance improvements over traditional work distributions. © Springer-Verlag Berlin Heidelberg 2007.