Reducing the Power Consumption of Matrix Multiplications by Vectorization
Authors:
- Thomas JakobsTJThomas Jakobs
- Michael HofmannMHMichael Hofmann
- Gudula RungerGRGudula Runger
- 1 August 2016
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 213-220
Abstract
The power consumption of programs and algorithms is currently a very active research field. This includes the investigation of the effect of different programming techniques on power consumption. Some programming techniques have already been studied intensively. However, there are techniques that did not get as much attention as needed so far. One of these techniques is the vectorization of programs, which uses special operations to calculate several data in one step. In this article, we investigate the effect of vectorization on the power consumption and study several program versions of dense matrix multiplication which combine vectorization with other techniques, such as loop unrolling or compiler options. We show that the use of vectorization is not only capable of improving the performance but can also reduce the power and energy consumption of programs.Keywords
This publication has 4 references indexed in Scilit:
- Modeling and analyzing the energy consumption of fork-join-based task parallel programsConcurrency and Computation: Practice and Experience, 2014
- Improving Energy Efficiency through Parallelization and Vectorization on Intel Core i5 and i7 Processors2012
- Can traditional programming bridge the Ninja performance gap for parallel computing applications?2012
- Code transformations and SIMD impact on embedded software energy/power consumption2009