Research on CUDA-based Kriging Interpolation Algorithm

摘要

Three-dimensional geological model can describe the types of geological information efficiently, express a variety of topological relations among geological phenomena intuitively. Kriging interpolation algorithm is an important spatial interpolation method of three-dimensional geological modeling, but every grid point needs to compute augmented matrix and solve equations, so it costs too much time. With the modeling scale and drilling data quantity' increasing rapidly, we need to make Kriging algorithm parallel imperatively. However parallel machine has many disadvantages- high cost for example, we propose to use CUDA which is multi-threaded parallelism and low lost GPU-using to implement Kriging interpolation algorithm in this paper, then optimize the memory access and so on according to the executive features of CUDA. We make high speedup and low running time. Compared with the single-threaded CPU implementation, the speedup of multi-threaded CUDA implementation achieves more than 110.