High-performance Video Decoding using Graphics Processing Units

Auteurs

• Biao Wang

Paramètres

• Informatique, programmation

En savoir plus sur le livre

The increasing demand for high-quality video decoding poses significant computational challenges for traditional Central Processing Unit (CPU) architectures. Graphics Processing Units (GPUs) generally offer superior computational power, but efficient GPU execution hinges on achieving massive parallelism and minimal execution divergence—criteria not met by all video decoding kernels. This research investigates the effective utilization of GPUs in video decoding applications, addressing challenges such as workload distribution between CPU and GPU, task optimizations across these heterogeneous devices, and their efficient communication. A complete parallel HEVC decoder was developed for CPU+GPU systems, leveraging available decoding parallelism on both devices. Two workload balancing schemes were implemented to adapt to variations in computational resources. An energy measurement module was also created for efficiency analysis. Results indicated that suitable decoding kernels can be accelerated significantly (up to 28.2×) on GPUs. At the application level, GPU architecture provides substantial acceleration with a limited number of CPU cores (1 to 8). For instance, on a system with an NVIDIA Titan X Maxwell GPU and an Intel Xeon E5-2699v3 CPU using four cores, the proposed HEVC decoder achieves 167 frames per second for 4K videos, yielding a 2.2× speedup over the state-of-the-art CPU decoder. However, when more than eight CPU core

Achat du livre

• Année de publication: 2018