Profile driven schemes for energy-sensitive cache hierarchy

Abstract

With the advent of mobile and handheld devices, power consumption in embedded systems has become a key design issue. Of the components that consume significant amounts of power in an embedded system, cache memories have been reported to consume in excess of 40% of the total power in typical high end embedded processors [38]. Therefore, cache memories are an obvious target of many low-power optimizations. Recently, it has been shown that cache requirements of the applications vary widely [74] and a significant amount of energy spent in cache accesses can be saved by tuning the cache parameters according to the needs of the application[60][62]. However, tuning the cache memory to suit the needs of the application entails identification of optimal cache configurations in the first place. With the large set of configurations to choose from, this process is prohibitively time consuming if done through exhaustive cache hierarchy simulations. Therefore, there exists a need for tools that can rapidly identify optimal cache configurations to tune the cache parameters for any given application. In this research work, we present novel, lightweight and fast techniques for energysensitive tuning of the instruction cache hierarchy. The proposed techniques rely on profiling the application to identify its loop characteristics. The loop characteristics are then used to identify the optimal instruction cache size for any application. The techniques are initially proposed for single-task based systems. Subsequently, they are extended to rapidly identify optimal instruction cache size for multitasking applications too. Cache tuning for RTOS-driven multitasking applications is achieved by intelligently separating the user tasks and RTOS components and profiling them in isolation to identify the nature of loops in them. We apply the proposed techniques to tune a predictor based filter cache hierarchy for instructions for both single-task based applications and RTOSdriven multitasking applications. The proposed techniques are able to identify optimal or 10 ATTENTION: The Singapore Copyright Act applies to the use of this document. Nanyang Technological University Library