Joint application-architeture design space exploration of multimedia applications on many-core platforms - an experimental analysis

The design space of mapping multimedia applications on an architectural platform is complex and many parameters are needed to be considered in order to find the optimum mapping. Conventionally, architectural parameters are varied to find different design points and the application side parameters are considered fixed and unchanged, while multimedia applications are equipped with several configurations to control the complexity and quality of the output. In this paper, we experimentally investigate joint application-architecture design space exploration of multimedia applications on many-core platforms. The joint exploration is conducted on two state of the art video coding standards and STHorm many-core platform as the underlying architecture. In the first case study, MPEG4-SP decoder is mapped on STHorm with varying buffer size and variable number of Processing Elements (PEs). In the second case study, HEVC (High Efficiency Video Coding) decoder with variable Quantization Parameter (QP) is mapped on STHorm with variable PE number. The application is characterized with representative parameters on the basis of high level dataflow representation of the application. It is demonstrated that joint exploration of these parameters outperforms that of their separate exploration in terms of well-established combinatory metrics such as space-time product and energy-time product metrics while achieving performance constraints.