A Real-Time H.266/VVC Software Decoder

Abstract

The new Versatile Video Coding Standard (VVC) was finalized in July 2020. This new international standard is able to provide a higher compression efficiency by up to 50% for the same subjective quality compared to its predecessor HEVC but at the cost of an increased computational load. This paper investigates the complexity of VVC decoder processing blocks and presents a highly optimized decoder implementation that can achieve 4K 60fps VVC real-time decoding on an x86 based CPU using SIMD instruction extensions of the processor and additional parallel processing including data and task-level parallelism.

Overview of the Versatile Video Coding (VVC) Standard and its Applications

“…Within a few months after the VVC standard was finalized, Fraunhofer HHI made their optimized Versatile Video Decoder (i.e., VVdeC) available on GitHub [73], [74]. Tencent also presented an independently developed VVC decoder (i.e., O266dec) [75] which was implemented from scratch and runs on multiple operating systems including Linux, Microsoft Windows, Apple Mac OSX, iOS and Android [76]. Alibaba demonstrated Ali266, a thin VVC decoder that runs efficiently on mobile phones of wide-ranging hardware capabilities [72].…”

“…They both aim at enabling live decoding of 4K video on modern x86 processors by incorporating:  Single-instruction multiple data (SIMD) architecture optimization of sample operations using instruction sets such as SSE42, AVX2 and AVX512, where applicable. It is observed that this optimization is specifically beneficial for the inverse transforms, motion compensation interpolation, and in-loop filters [73], [75].  Multi-threading at the picture level, CTU level, task level and sub-CTU level.…”

“… Multi-threading at the picture level, CTU level, task level and sub-CTU level. More detail about multi-threaded implementations is provided in [73] and [75].…”

“…In addition, O266dec used bit depth templatization in its design to allow one unified decoder implementation to easily support various input bit depths without performance penalties [75], [76], in contrast to always using a 16-bit internal memory structure as in the VTM reference software.…”

VVC Search Space Analysis Including an Open, Optimized Implementation

“…Measured in terms of runtime, in the commonly used random access configuration, VVC software test model VTM takes around 8× more time to encode and 2× more to decode than the HEVC software test model, HM [5]. First reports of optimized implementations including real-time decoder implementations [6][7][8] and optimized encoding algorithms [9][10][11][12][13] are available with very improved runtime. Those indicate that the VVC complexity, and its complexity increase when compared to HEVC, can only in limited manner be measured when comparing the reference software runtimes.…”