Towards Exploiting Intra-Application Dynamism using an H.264 Codec

The computational requirements of applications targeting multiand manycore systems increase rapidly. In order to achieve optimized execution, streaming applications must dynamically adapt to changes of their input. In this work, this intra-application dynamism is explored using an H.264 standard complying Codec pair. Initially, an encoder/decoder pair of applications is implemented for the DAL model of computation, using code targeting the HOPES framework as baseline. Focus is given on the encoder application, where we apply the wavefront parallelism technique in order to increase the degree of parallelism. This degree is de ned as a function of the input video's resolution, which introduces static dynamism to the application. As a means of increasing the encoder's performance, a frame pipelining scheme is proposed. In addition, we argue that run-time dynamism can be achieved by balancing the tradeo between execution time and compression rate. Three encoding parameters are proposed that can be adjusted at run-time to achieve run-time dynamism. Finally, the implementation is evaluated, by calculating the maximum theoretical speedup that it can achieve and by exploring the e ects of di erent mappings and di erent degrees of parallelism on the performance. The results show that we have achieved better maximum speedup than the HOPES implementation.