GPU-based Parallelization Algorithm Research For 2D Line Integral Convolution

Two-dimensional flow visualization is one of the important issues in the ocean information visualization research. With the dramatic development of science and technology, the ocean vector field data collected is very large, so single calculation can not meet the requirements of large-scale computing. Therefore, high-performance parallel computing has become the only way to perform the real-time visualization of massive data.Recently the performance of the graphics hardware is promoted greatly, and shows great possibilities to be as a general purpose processing units as well as been the tool for graphics manipulation. Particularly, after the nVIDIA company launched Compute Unified Device Architecture (CUDA) platform, GPU is able to solve complex computational problems quickly, so it has entered upon the high computing parallelism. GPU performance computing has become the global trend in high-performance computing and indicated an important direction in the field of marine environmental information.Line integral convolution method (LIC) is an important method in the vector field visualization, which shows the integrity and global characteristics of the marine environment and reflects the structure of the vector field. It can transform the ocean flow field image data to graphics and images, which can display on computer screen for getting the information of large scale ocean flow structure by human-computer interaction. But in the current study, we often face a major problem, which is the algorithm needs considerable computational time. So it is difficult to achieve the effect of real time visualization. This paper will present a GPU-based Line Integral Convolution parallel algorithm for visualization of discrete vector fields to accelerate Line Integral Convolution algorithm. The algorithm is implemented with parallel operations using Compute Unified Device Architecture programming model in GPU. The method can provide up to about 50×speedup without any sacrifice on solution quality, compared to conventional sequential computation. In order to reflecting the GPU's superiority in parallel computing on real-time visualization of large-scale marine data, in this paper, cluster based Line Integral Convolution parallel algorithm is used to be a reference. Through the comparison of both computing preference, GPU based parallel algorithm shows a higher degree of parallelism and less computation time in the large scale vector field visualization. Experimental results show that when the parallel algorithm based on two different mechanisms maintain the same characteristics of the vector field, GPU based parallel algorithm is better, improve the processing time more effectively and get to the purpose of real-time visualization. It provides an important new approach to parallel deal with large-scale planar flow field.