Influence of software optimization on energy consumption of embedded systems

Abstract

The main source of power consumption in a digital system is dynamic power dissipation. The chapter shows that program optimization has the positive influence on power consumption. The system level optimization has the greatest effect on potential power consumption gains. The chapter is focused on the transformations of program loops as the point where the most of computational load exists. Some optimized and parallelized software are analyzed from the point of power consumption. These results show the influence of program optimization on the power consumption and possibility of high-quality low-power design of embedded systems. The loop fusion algorithm for programs optimization is presented and its influence to the power consumption is shown. The experiments show that the loop fusion optimization may decrease the current consumption by more than 20%. The real applications are considered as examples of embedded systems use. A program of contour extraction in medical images is considered as the example of usage of multidimensional loops fusion algorithm. The example shows that the decrease in the activity of elements leads to reduction of energy consumption. The authors propose the loop fusion method for high level language code-to-code transformations and demonstrate its efficiency in terms of power consumption.