摘要
GPU/CPU协同并行计算是一项全新的、具有革命性的技术,将对地震资料处理行业产生深远的影响。本文在充分发挥GPU存储带宽宽、多寄存器和多处理器结构特点的基础上,通过改进地震成像并行计算模块的核心函数,形成一套波动方程地震成像GPU高效并行算法,并取得理想的加速比和可扩展性。所形成的一套GPU/CPU平台下波动方程叠前深度偏移软件,已投入地震资料处理实际应用。理论模型试算和实际资料的处理验证了本方法的有效性和实用性。
GPU/CPU Co-Processing Parallel Computing(GPPC) is a new revolutionary technology,which will make great contributions on seismic data processing.Based on the hardware structural characters of Graphic Processing Unit including wide memory band,multi-registers and multi-processors,this paper introduces a wave equation imaging parallel method with high efficiency by improving on the kernel function of seismic imaging parallel calculation modules,and produces ideal speed-up ratio and expansibility.The study develops one wave equation PSDM technology and software based on GPPC,which have been applied in seismic data processing.And the tests on the SEG/EAGE salt model and 3D real seismic datasets prove the usefulness and practicability of this method.
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2011年第5期705-709,836+661,共5页
Oil Geophysical Prospecting
基金
中石化前瞻性课题"地震叠前成像GPU并行计算技术研究"
关键词
图形处理器
统一计算设备架构
波动方程叠前深度偏移
Graphic Processing Unit,Compute Unified Device Architecture,wave equation prestack depth migration