Complexity Analysis of a Versatile Video Coding Decoder over Embedded Systems and General Purpose Processors

Abstract

The increase in high-quality video consumption requires increasingly efficient video coding algorithms. Versatile video coding (VVC) is the current state-of-the-art video coding standard. Compared to the previous video standard, high efficiency video coding (HEVC), VVC demands approximately 50% higher video compression while maintaining the same quality and significantly increasing the computational complexity. In this study, coarse-grain profiling of a VVC decoder over two different platforms was performed: One platform was based on a high-performance general purpose processor (HGPP), and the other platform was based on an embedded general purpose processor (EGPP). For the most intensive computational modules, fine-grain profiling was also performed. The results allowed the identification of the most intensive computational modules necessary to carry out subsequent acceleration processes. Additionally, the correlation between the performance of each module on both platforms was determined to identify the influence of the hardware architecture.

GPU-based parallelisation of a versatile video coding adaptive loop filter in resource-constrained heterogeneous embedded platform

“…For instance, the inclusion of an Adaptive Loop Filter (ALF) in the filtering loop is responsible for reducing the coding artefacts and minimising the mean square error between the original and reconstructed samples. However, these in-loop filters significantly increase the computational complexity requirements of a decoder, on average 30% and 40% when decoding on a high-performance general purpose processor (HGPP) and an embedded general purpose processor (EGPP), respectively [4,5]. According to [5], ALF alone represents an average computational complexity of 5-12% and 12-24% of the total decoding time share on an HGPP and an EGPP, respectively.…”

“…However, these in-loop filters significantly increase the computational complexity requirements of a decoder, on average 30% and 40% when decoding on a high-performance general purpose processor (HGPP) and an embedded general purpose processor (EGPP), respectively [4,5]. According to [5], ALF alone represents an average computational complexity of 5-12% and 12-24% of the total decoding time share on an HGPP and an EGPP, respectively. As a result, it has become a rather challenging research goal to reduce the in-loop filtering time, and more particularly to reduce the ALF filtering time through parallelisation to achieve realtime decoding.…”

Gpu-Based Parallelisation of a Versatile Video Coding Adaptive Loop Filter in Resource-Constrained Heterogeneous Embedded Platform

“…For instance, the inclusion of an Adaptive Loop Filter (ALF) in the filtering loop is responsible for reducing the coding artifacts and for minimising the mean square error between the original and reconstructed samples. Nonetheless, these in-loop filters significantly increase the computational complexity requirements of a decoder, on average 30% and 40% when decoding on a high-performance general purpose processor (HGPP) and embedded general purpose processor (EGPP), respectively [4], [5]. According to [5], ALF alone represents an average computational complexity of 5%-12% and 12%-24% of the total decoding time share on a HGPP and on a EGPP, respectively.…”

“…Nonetheless, these in-loop filters significantly increase the computational complexity requirements of a decoder, on average 30% and 40% when decoding on a high-performance general purpose processor (HGPP) and embedded general purpose processor (EGPP), respectively [4], [5]. According to [5], ALF alone represents an average computational complexity of 5%-12% and 12%-24% of the total decoding time share on a HGPP and on a EGPP, respectively. As a result, it has become a rather challenging research goal to reduce the in-loop filtering time, and more particularly to reduce the ALF filtering time through parallelisation in order to achieve real-time decoding.…”

Fusion-Based Versatile Video Coding Intra Prediction Algorithm with Template Matching and Linear Prediction

“…Wang et al [33] designed a Sample Adaptive Offset (SAO) acceleration method to reduce the complexity of VVC. Saha et al [34] analyzed the decoder complexity of VVC on two different platforms.…”