VC-Bench: A Video Coding Benchmark Suite for Evaluation of Processor Capability

Video coding standards such as H.264 are widely used since the flourish of digital compressed video ranging from low bit-rate Internet streaming applications to HDTV broadcast. As a result, more powerful processors are required to meet the growth in coding quality and velocity. Though many efforts have been made on associated items, it lacks one up-to-date benchmark suite for studying architectural properties of video coding applications. This paper introduces VC-Bench, a Video Coding Benchmark suite which is built up from a wide range of video coding applications. Firstly, typical codecs are selected (X264, XVID, VP8) according to the popularity, coding efficiency, compression quality, and source accessibility. Secondly, hotspots are extracted from coding process as kernels. VC-Bench mainly focuses on Transformation, Quantization and Loop filter. All of the extracted kernels are single-threaded version without any architecture-specific optimizations, such as SIMD. Besides, inputs in three different sizes are provided to control running time. Finally, to better understand intrinsic characteristics of video coding application, we analyze both computational and memory characteristics, and further provide insights into architectural design which can improve the performance of this kind of applications.