Fine-grain instruction scheduling for low energy

Energy consumption is increasingly becoming an important metric in designing computing systems. This paper presents an instruction scheduling algorithm to reduce energy consumption in processor data-paths. The unique aspect of our algorithm is that it is fine-granular; i.e., it works on a pipeline stage granularity. This is in contrast to current energy-aware instruction scheduling techniques that work on an instruction granularity. Our preliminary experimental results indicate that our fine-granular approach both leads to schedules with lower energy consumption (as compared to coarse-grain techniques) and helps us estimate the absolute data-path energy consumed by the code better.