查看文章

With the rapid development of the Internet of Things, tensor-based coding and decoding algorithms are widely used in wireless camera networks. Recently, a novel video decoder based on the low-tubal-rank tensor model has been proposed, which achieves better quality of services than conventional schemes. However, the tensor decoding algorithm is compute-intensive, rendering it impractical for real-time applications. In this paper, we propose effective strategies to accelerate the tensor decoder on GPUs (Graphics Processing Units). We implement the tensor decoding algorithm on the GPU architecture, and propose optimization strategies to eliminate data reorganizing overhead, provide batched complex matrix computations, and reduce memory consumption as well as computation overhead. With real video data, the GPU algorithm achieves an average of 237.44× and up to 312.39× speedups on a Tesla …