Realization of power aware software prefetching as a multi-objective optimization problem

Software prefetching is a performance-oriented optimization technique, which is generally used to reduce the gap between processor speed and memory access speed. When software prefetching is applied to memory-intensive benchmark programs, the performance improves with higher power consumption. The present work provides a mechanism to transform a program with software prefetching to its power-aware equivalent. This is done by executing the software prefetching program at different voltage-frequency pairs. Besides reducing the power, the performance has been improved by adjusting the prefetch distance. XEEMU-Panalyzer simulator is used to evaluate the present work. Experimental results of the proposed scheme guarantees that performance improvement of software prefetching program is possible at the cost of less power consumption. The proposed work can enable a compiler to generate power aware software prefetching program.