Performance and Energy Consumption Characterization and Modeling of Video Decoding on Multi-core Heterogenous SoC and their Applications

To meet the increasing complexity of mobile multimedia applications, the System on Chip (SoC) equipping modern mobile devices integrate powerful heterogeneous processing elements among which General Purpose Processors (GPP), Digital Signal Processors (DSP), hardware accelerator are the most common ones. Due to the ever-growing gap between battery lifetime and hardware/software complexity in addition to application computing power needs, the energy saving issue becomes crucial in the design of such systems. In this context, we propose a study aiming to enhance the understanding of the energy consumption behavior of video decoding on these kinds of systems. Accordingly, an end-to-end methodology for characterizing and modeling the performance and the energy consumption of video decoding on GPP and DSP is proposed. The characterization step is based on an exhaustive experimental methodology for evaluating, at different abstraction levels, the performance and the energy consumption of video decoding. It was achieved on embedded platforms on which were executed a wide range of video decoding configurations. This step highlighted the importance to consider different parameters which may pertain to different abstraction levels in evaluating the overall energy efficiency of a given system. The measurements obtained in this step were used to build empirically performance and energy models for video decoding on both GPP and DSP. The proposed models gave very accurate estimation (R2 = 97%) of both the performance and the energy consumption of video decoding in terms of a rich set of parameters including the video quality and the processor frequency. Moreover, based on a multi-level characterization and sub-model decomposition approaches, we show how the developed models, unlike classic empirical models, are easily and rapidly generalizable to other platforms. Some possible applications using the developed models, in the context of adaptive video decoding, were proposed. In general, it consists to use the capability of the proposed performance model to predict the decoding time of a given video quality in dimensioning/scheduling the processing resources. Due to the increasing demand on High Definition (HD), the characterization methodology was extended to consider HD video decoding on both parallel multi-cores and hardware video accelerator. This part highlighted the potential of parallelism video decoding to increase the energy efficiency of video decoding and point out some open issues in this domain.