Energy-Aware Loop Parallelism Maximization for Multi-core DSP Architectures
Meikang Qiu, Jianwei Niu, Laurence Tianruo Yang, Xiao Qin, Senlin Zhang, Bin Wang · 2010
With the advance of semiconductor, multi-core architecture is inevitable in today's embedded system design. Nested loops are usually the most critical part in multimedia and high performance DSP (Digital Signal Processing) systems. Hence, maximizing loop parallelism is an important issue to improve the performance of a modern compiler. This paper studies how to maximize the system performance with the consideration of energy reduction for applications with multidimensional nested loops on multi-core DSP architectures. An algorithm, EALPM (Energy-Aware Loop Parallelism Maximization), is proposed in this paper. We implemented a two phase strategy. First, the strategy uses retiming and loop transformation to parallelize nested loops. Then the strategy employs a novel voltage assignment algorithm to reduce total energy consumption. The experimental results show that on average, both performance and energy-saving can be significantly improved using the EALPM algorithm.