| The research contents in this paper,, come from the seismic reservoir predictionsystem of company,ROCKSTAR,GeoScope in English. The software is mainly used todealwith pre-stack, post-stack data.It contains multiple subsystems,absorbtion moduleis a basic module of GeoScope. And in the petroleum industry has been widelypromoted and applied. With the growing of the software and the requirements ofsoftware real-time of client are more and more higher. The problem is mainly limitedby the size of the data processing method, and constraints.Data size can usually be divided into ultra-large-scale, large-scale and generalscale.But the scale of the absorbtion module is rather awkward,because it is betweenlarge and general scale, typically hundreds of G single-precision data. To solve theselarge data, people usually will use supercomputers,and this is where the problem,cost control and efficiency control,comes from. Especially in the absorptionmodule,this contradiction becomes even more acute.June2007, NIVIDA introduced CUDA (Compute Unified Device Architecture)technology,combined with graphics processing techniques and parallel programmingtechniques.format a called CUDA programming language,similar to C-style, makingthe solution to this conflict no longer so difficult. It is a heterogeneous programmingmode, take full advantage of the massively parallel GPU computing capacity andCPU’s logic computing power.The problem will be divided into two parts: the serialone and parallel one,and then to resolve them separately with CPU and GPU. Thiswill not only improve the petroleum industry exploration data processing efficiencybut also reducing costs.The research works in the paper, which is based GeoScope platform, optimize the absorption module in the subsystem. The specific contents include the following:1. The absorption module backup material;2. The CUDAtechniques;3. The achievement of absorption module with CUDA;... |