Exploiting program phase behavior for energy reduction on multi-configuration processors

Energy consumption is a major design issue for modern microprocessors. In previous work, several techniques were presented to reduce the overall energy consumption by dynamically adapting various hardware structures. Most approaches however lack the ability to deal efficiently with the configuration space explosion in case of multiple adaptive structures.In this paper, we present a mechanism that is able to deal with this configuration space problem. We first identify phases through profiling using a new phase classification method and determine the optimal hardware configuration per phase using an efficient offline search algorithm. During program execution, we inspect the phase behavior and adapt the hardware on a per-phase basis. Using SPEC2000 benchmarks, we show that the proposed mechanism achieves an energy reduction of 36% on average with an average performance degradation of only 2.9%. We also show that online processor configuration optimization is far less effective for multi-configuration processors, with an average energy reduction of less than 20% for comparable performance degradations.