Fine-grain compensation method with consideration of trade-offs between computation and data transfer for power consumption

Abstract

Fine-grain parallelizing method with consideration of the number of data transfers for low power consumption is proposed. In the proposed method, power consumption by data transfers between processor elements in a multiprocessor is focused on, and the number of data transfers is reduced. In this paper, a measure based on the relationship between variables in a given program is defined to evaluate the number of data transfers, firstly. And then a proposed compensation method by use of the evaluation of power consumption based on the measure is explained. Finally, the result of applying proposed compensation method implemented on COINS framework to several example programs is shown.