GPU-based Parallel Discrete Cosine Transform Algorithm

With the social progress and the development of science and technology, People need to deal with the increasing scale of the data and more complex problems,but single CPU can not meet these requirements. so parallel computing cames out and play more and more important part in this world.GPU, As part of parallel computing, because of its powerful data-parallel processing capability,it is widely used in digital imaging and other fields.In the field of digital image processing, discrete cosine transform (DCT) is theoretically considered to be a good way, because of its good energy compaction properties and fast algorithms,it is widely used in such as JPEG, MPEG image compression standards and so on. Two-dimensional discrete cosine transform algorithm is widely used in image processing, but the complexity of the algorithm can be achieved o(n4), with the transformation of image data increases, the processing time will become very long. This paper change the serial algorithm into parallel algorithm, running the parallel algorithm on the GPU and won more than times speedup. In addition, this paper proposes two different GPU-based DCT algorithm, and analyzes the factors that affect the GPU speedup to achieve how to write a more efficient parallel algorithm for the goal.First,the paper analyses the parallel computing system architecture, programming model, and some of the parallel algorithm performance evaluation methods. Second,the paper analyses the hardware architecture and the CUDA programming model, and proposed method of CUDA programming steps.And then,the paper analyses the DCT in JPEG and MPEG image compression standard application.At last,this paper implements the serial two-dimensional DCT algorithm and GPU-based two-dimensional DCT algorithm. In this paper, two different GPU-based two-dimensional DCT algorithm, relative to the two-dimensional DCT of the serial algorithm, these two parallel algorithms to obtain a high speedup.