A Measurement-Based Automatic Energy Optimization Technique for Embedded Applications

摘要

Low-energy consumption, high-performance, and small code size are the three most important requirements in designing modern embedded applications. In this paper, we describe a measurement-based automatic energy optimization technique for embedded applications. To realize this goal, we have accomplished two major tasks as the major contributions of this work. First, we design and implement a novel embedded development tool, called ePRO (energy profiler and optimizer), which measures and analyzes the energy consumption, performance, and code size of embedded applications in an integrated fashion. Second, we propose and implement a measurement-based optimization technique which is mainly based on the compiler-level optimization-space exploration (CL-OSE) as an internal optimization engine of ePRO. Over the repeated experiments, we have verified that the proposed technique improves the energy consumption of various embedded applications by 16.8% over the most aggressively compiled executables while maintains the time overhead within a reasonable level.