HEVC video decoder acceleration using GPU compute
The video decoder is an integral part of consumer devices such as Mobile Phones, Tablets, Set Top Box and Smart TVs. Video Processing (decode/encode) is generally performed by the Hardware Accelerators in the SoC or with Software decoding on the powerful multi-core Cortex-A series CPU.
Most of today’s processors are equipped with powerful GPUs that support massive data parallelism, are optimized for graphics processing and have compute ability that surpasses the capability of CPU’s. However the GPU’s architecture makes it unsuitable for video decoding and hence this power has typically remained unharnessed for Video Processing.
By identifying operations that are highly parallel and efficiently offloading these to GPUs, a significant improvement in performance is achieved i.e., a higher FPS throughput for the same CPU usage. Alternatively, for the same performance throughput, GPU acceleration reduces CPU load significantly, resulting in lower power consumption since GPUs are more power efficient for compute algorithms.