Abstract: The information industry’s energy consumption is increasing with the improvement of the information technology. Especially higher chip integration, more complex system applications make power consumption problem become a key issue. Pure hardware power optimization can no longer meet the further improvement of the power optimization requirements, thus software-based power optimization has achieved good results. This paper reduces the power of the system by reducing the number of bus flip during the compilation process. For instruction address bus, it allocates the function segment by using genetic algorithm. And then it uses T0 code to reduce the number of bus clip. For data bus, it uses the ant colony optimization to schedule the instructions and uses 0-1 invert code to reduce the number of the bus clip. It designs and implements the low power optimization algorithms for data bus and address bus, and verifies the algorithm by experiences and it can reduce about 25% of the bus power.

Key words: low power, compiler optimization, bus coding, instruction schedule

摘要： 随着社会信息化水平的不断提高，信息产业的快速发展，由此带来了能源的消耗也越来越高。特别是芯片集成度越来越高，系统应用越来越复杂，这就使得功耗问题成为嵌入式系统必须面对的一个关键问题。单纯的硬件功耗优化已经不能满足要求，基于软件的功耗优化取得了很好的成效。在编译阶段，通过减少总线的翻转次数来降低系统的功耗。针对指令地址总线，结合遗传算法进行函数段的分配，结合相关的编码策略，减少总线翻转，从而降低其功耗。针对数据总线，使用蚁群算法进行指令调度，用0-1翻转编码，有效减少了其总线翻转，降低了功耗。这种基于数据总线和地址总线的优化算法，能够在特定的实验平台下通过实验验证，算法对于总线功耗的优化效率大约为25%左右。

关键词: 低功耗, 编译优化, 总线编码, 指令调度