Green Demands Oriented Data Allocation for Embedded Systems

Green demands, such as energy efficiency and resource utilization, have become critical issues during the design of embedded systems. Data allocation, one of the most important back-end optimization methods of compiler, can largely influence the utilization of energy and resources. This paper proposes a data allocation approach to improve the effective utilization of resources and energy. First, a green evaluation model for data allocation is proposed in this paper. In this model, green indicators are proposed to represent both energy efficiency and resource utilization. Second, based on the evaluation model, an iterative-style multi-objective data allocation approach is proposed to reduce the energy consumption and to balance the resource utilization. This data allocation approach resorts to two common compilation optimization techniques, i.e., exchangeable instructions rearrangement and register reallocation, to improve the green indicators. In addition, an iterative framework is employed to synthesize the exchangeable instructions rearrangement and register reallocation techniques smoothly to improve the green indicators further. Simulation experiment results show that the proposed method can obtain about 23% improvement on average when GCC compiler is the baseline. Therefore, the proposed method can significantly improve the green indicators.