Collaborative SMA: Dynamic Program Phase Adaptation and Reconfiguration in Multiprocessor Systems

Cornell University, Ithaca NY

Investigators

Abstract

The goal of the project is to create a framework to enable the design of reconfigurable multiprocessors . This project addresses and develops solutions for the problems of: (a) dynamic identification and prediction of program phases in parallel codes, and (b) their application to reconfigurable parallel computing. The proposal will: (1) Study the design and implementation of multiprocessor phase identifiers and their integration with phase predictors, and specifically will analyze the shortcomings of current uniprocessor approaches in the context of multiprocessor systems, and propose alternative/supplementary means of gathering and utilizing phase information. (2) Develop novel metrics suitable for quality assessment of multiprocessor phase identifiers and predictors overcoming the shortcomings of traditional uniprocessor approaches. (3) Aggregation and integration of COTS phase-adaptive systems (e.g., adaptive chip multiprocessors). (4) General exploration of phase-adaptive multiprocessor systems.