【摘要】 地震波的叠前时间偏移算法是构造复杂岩层成像最有效的方法之一。地震勘探进入海量数据时代,且叠前偏移算法是数据处理中最费时的环节,对叠前偏移算法做并行计算优化有着重要的研究意义。近年来,高性能并行计算开始进入异构、众核时代,以Intel新一代至强融核MIC(Xeon Phi)为例,新型众核处理器具有成本低、性能高等特点。从最经典的Kirchhoff叠前时间偏移(PKTM)算法出发,基于CPU+MIC异构平台,采用offload编程模式实现对PKTM算法的并行移植与性能优化,对于6 000万规模(8 000×8 000)的应用问题,总的并行模拟时间从357.52s减少到1.66s,性能提升了214.37倍。更多还原

【Abstract】 An efficient technique which is now being implemented in photographing images of complicated rock stratum is the seismic wave PKTM algorithm.With the earthquake prediction coming into massive data generation,it is of essential importance to optimize this algorithm by parallel computation.In recent years,high performance parallel computation is characterized by heterogeneous and many cores systems.A typical example of this kind of processors,featured with low cost and high performance is Xeon Phi,being known as MIC.On the basis of the classic PKTM algorithm,we parallelize and optimize the PKTM algorithm in the offload programming model,based on CPU+MIC heterogeneous platform.For applications with the scale of 64 000 000(8 000×8 000),the total parallel simulation time is reduced from 357.52 seconds to 1.66 seconds,achieving 214.37 xperformance improvement.更多还原