基于GPU的吸收模块并行化改进

发布时间：2018-06-09 11:33

本文选题：吸收模块 + CUDA　；参考：《成都理工大学》2013年硕士论文

【摘要】：论文的研究内容来源成都晶石石油科技有限公司所研究的地震储层预测系软件，GeoScope，该软件主要用来处理叠前、叠后数据。它包含多个子系统，其中吸收模块就是该软件子系统中的一个基础模块。并且在石油行业内得到了广泛的推广和应用。随着软件不断的壮大和客户对软件实时性的要求越来越高，软件的效率问题逐渐被放大。该问题主要受限于处理方法和数据规模的限制。数据规模通常可以分为超大规模，大型和普通三种。而吸收模块所处理数据的规模比较尴尬，处在大型和普通型之间，通常是数以百G的单精度数据。用巨型机或者超级计算机解决大规模的问题时会面临成本控制和效率控制这样的矛盾。特别是当吸收模块处理这种数据的时候，这个矛盾就变得更加尖锐。 2007年6月，NIVIDA公司推出了CUDA(Compute Unified DeviceArchitecture，统一计算设备架构)技术，结合图形处理学技术，将用于巨型机和超级计算机中的并行编程技术进行封装，形成了名为CUDA的类C风格编程语言，使得这个矛盾的解决变得不再那么困难。它是一种异构编程模式，充分利用GPU大规模并行计算的能力和CPU的逻辑计算能力。把问题分解为串行和并行两个部分，用CPU和GPU各自的优势分开解决。这样不仅提高了石油行业内勘探数据处理效率的同时，，也降低了成本。本论文所进行的研究工作，是基于GeoScope平台，对其子系统中的基础模块吸收模块进行了研究和并行优化处理。具体内容包括： 1.吸收模块背景技术资料； 2. CUDA并行编程技术； 3.使用CUDA技术对吸收模块并行化改进的具体实现。
[Abstract]:The research content of this paper comes from GeoScopesoftware, a seismic reservoir prediction software developed by Chengdu Jingshi Petroleum Technology Co., Ltd., which is mainly used to process prestack and post-stack data. It consists of several subsystems, and the absorption module is one of the basic modules in the software subsystem. And in the oil industry has been widely promoted and applied. With the development of software and the requirement of software real-time, the problem of software efficiency is magnified gradually. This problem is mainly limited by the processing method and the data scale, which can be divided into three types: large scale, large scale and common scale. The size of the data processed by the absorption module is awkward, usually hundreds of gigabytes of single-precision data between large and common types. Using supercomputers or supercomputers to solve large-scale problems will face such contradictions as cost control and efficiency control. Especially when the absorption module processes this kind of data, the contradiction becomes more acute. In June 2007, NIVIDA introduced the CUDAT Compute Unified device Architecture (Unified Computing device Architecture) technology, combined with graphics processing technology. The parallel programming technology used in supercomputers and supercomputers is encapsulated to form a C-style programming language called CUDA which makes the solution of this contradiction less difficult. It is a heterogeneous programming mode, which makes full use of GPU's ability of large-scale parallel computing and CPU's logical computing ability. The problem is divided into two parts, serial and parallel, and solved separately with the advantages of CPU and GPU. This not only improves the efficiency of exploration data processing in petroleum industry, but also reduces the cost. The research work carried out in this paper is based on GeoScope platform. The basic module absorption module in its subsystem is studied and the parallel optimization processing is carried out. The specific contents include: 1. Absorption module background technical data; 2. CUDA parallel programming technology; 3. Using CUDA technology to improve the parallelization of absorption module.
【学位授予单位】：成都理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P631.4;TP338.6

【共引文献】