FPGA与GPU并行计算分析——以Kirchhoff叠前时间偏移为例

Analysis of GPU and FPGA parallel computing——example of Kirchhoff prestack time migration

高性能计算(high performance computing)是地震数据处理领域一个重要研究课题.目前的高性能计算大多是利用图形处理器(GPU)强大并行处理能力,以GPU/CPU协同并行计算构架为基础实现有关算法.本文讨论了利用CUDA和Impulse C-to-FPGA工具分别在GPU(图形处理器)和FPGA(可编程逻辑门阵列)平台上并行实现Kirchhoff叠前时间偏移算法.处理结果表明,利用GPU加速Kirchhoff叠前时间偏移处理得到了14倍的单核加速比,利用FPGA加速Kirchhoff叠前时间偏移处理得到了9.5倍的单核加速比.

High performance computing is an important research subject in the area of seismic data processing. At present, most of the high performance computing is to use the powerful paralld processing ability of GPU and implement the algorithms on the basis of GPU/CPU co-processing parallel systerru The paper discusses parallel implementation of prestack Kirchhoff time migration algorithm on GPU（ Graphic Processing Unit ） using CUDA, as well as on FPGA（Field-Programmable Gate Array ） with the Impulse C-to-FPGA tool. Results show that GPU implementation achieves approximately 14 speedup, and FPGA implementation achieves about 9.5 speedup.