Development and optimization of high level dataflow programs: The HEVC decoder design case

With the standardization of the new High Efficiency Video Coding (HEVC) compression algorithm, a dataflow specification of the HEVC decoding process is also available as part of the standard. This paper presents methodologies to improve and optimize the performance of implementations derived by the dataflow specification. Regarding the architectural aspect of dataflow network, the throughput has been increased by developing more potential parallelism. For the platform aspect, critical processes have been optimized by applying SIMD functions and communications have been improved by cache efficient FIFO implementation. Results revealed an average acceleration factor of 7 in the decoding framerate over the reference dataflow implementation.