high performance computing on graphics processing units: hgpu.org

H.264/AVC is a recent video standard embraced by many multimedia applications. Because of its demanding encoding requirements, a high amount of computational effort is often needed in order to compress a video stream in real time. The intra-prediction and encoding are two of several modules included by H.264 that requires a high computational power. On the other hand, the GPU computational capabilities are exhibiting their supremacy for solving certain types of problems with data parallelism, which composes the majority of the intra-prediction and encoding process. This dissertation presents a parallel implementation of the intra-prediction and encoding modules by adopting a parallel programming API denoted by CUDA, which explores the massive parallelization capabilities of recent NVIDIA graphic cards in order to reduce the encoding time. The developed solution is integrated in an existing encoder, where the intra-prediction and the respective encoding are processed sequentially. Through the result of several conducted tests it is demonstrated that the developed module is capable to speed up the sequential execution beyond the computing capabilities of a recent CPU.