项目名称: 面向逆时偏移算法的FPGA加速技术研究
项目编号: No.61303003
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 付昊桓
作者单位: 清华大学
项目金额: 27万元
中文摘要: 近年来,石油勘探和开发开始转向深海等更复杂的地质区域。地震数据规模越来越大,处理算法也开始转向逆时偏移等更准确但同时更复杂的算法。相对于传统的Kirchhoff偏移算法,逆时偏移算法的计算量要高出一百倍以上,对现有计算平台的处理能力提出了巨大的挑战。相对于CPU和GPU等结构固定的处理器,基于FPGA的计算加速平台,可针对应用定制硬件结构,具有更广阔的性能优化空间。本项目将以硬件可编程的FPGA加速平台为基础,针对逆时偏移算法,实现软硬件的协同设计:(1)在软件算法上,针对FPGA的架构特点,研究和探索不同格式的有限差分算子;(2)在硬件架构上,研究和设计高并行度的数据处理流水线和高效缓存,以解除内存带宽对计算性能的限制;(3)研究计算精度优化技术,来进一步提升在固定带宽和固定硬件资源下可获得的计算性能。本项目将为FPGA在科学计算其他领域开展应用提供良好的参考和借鉴。
中文关键词: 可重构计算;FPGA;地震资料处理;大气模拟;
英文摘要: In recent years, the exploration of oil and gas is moving to more complicated geological areas, such as the deep sea. Therefore, the significantly-increaesd data volume and more complicated migration algorithms, such as the reverse time migration (RTM), bring tough challenges for the current high performance computers. Compared with CPUs and GPUs, which have a fixed architecture, FPGA-based accelerators provide the unique flexibility to customize the hardware architecture for a specific application, and more optimization options to improve the computation performance and to reduce the power consumption. Based on the reconfigurable hardware within FPGA accelerators, we try to achieve the co-design of both the software algorithm and the hardware architecture for RTM: (1) on the algorithm side, we explore different finite difference stencils to fit to the underlying FPGA architecture; (2) using the reconfigurable FPGA hardware, we study and design highly parallel data processing pipelines and customized buffer architectures so as to eliminate the memory bandwidth bottleneck; (3) we investigate the precision optimization methods to further improve the data throughput and the computation performance that can be achived within the given bandwidth and capacity of the FPGA devices.The research of this project would serv
英文关键词: reconfigurable computing;FPGA;seismic data processing;atmospheric simulation;