Energy-efficient stream task scheduling scheme for embedded multimedia applications on multi-issued stream architectures

With the increasing requirement of computation and data amount in embedded multimedia applications, stream architectures are exploited to accelerate the multimedia processing. The increasing arithmetic units on stream architectures make energy consumption a critical metric. In this paper, an energy-efficient stream task scheduling scheme is proposed for embedded multimedia applications on a multi-issued stream architecture. The proposed scheduling scheme takes four steps for task mapping, V/F scaling and record reordering. The target stream application is first mapped to the processor in pipeline to reduce task dependency. Then the supply voltage and frequency of tasks are scaled down for energy reduction by DVFS without violating real-time constraints. The computation intensities of tasks are taken into consideration as the priority for more efficient energy reduction. Besides, when records are dependent, more parallelism is exploited and records are reordered to improve the processing performance. The proposed scheme is evaluated with a set of benchmarks and 14% energy reduction is achieved compared with the GeneS algorithm.