Inferring Energy Bounds via Static Program Analysis and Evolutionary Modeling of Basic Blocks
The ever increasing number and complexity of energy-bound devices (such as the ones used in Internet of Things applications, smart phones, and mission critical systems) pose an important challenge on techniques to optimize their energy consumption and to verify that they will perform their function within the available energy budget. In this work we address this challenge from the software point of view and propose a novel parametric approach to estimating tight bounds on the energy consumed by program executions that are practical for their application to energy verification and optimization. Our approach divides a program into basic (branchless) blocks and estimates the maximal and minimal energy consumption for each block using an evolutionary algorithm. Then it combines the obtained values according to the program control flow, using static analysis, to infer functions that give both upper and lower bounds on the energy consumption of the whole program and its procedures as functions on input data sizes. We have tested our approach on (C-like) embedded programs running on the XMOS hardware platform. However, our method is general enough to be applied to other microprocessor architectures and programming languages. The bounds obtained by our prototype implementation can be tight while remaining on the safe side of budgets in practice, as shown by our experimental evaluation.