| 摘 要: 为提高叠前逆时偏移计算效率，本文采用MPI+CUDA混合粒度相结合的并行模式，对地震数据进行数据 分割，合理划分并行任务。总结出MPI+CUDA并行编程模型，提出叠前逆时偏移的混合粒度并行算法。根据CUDA特 有的存储方式，对叠前逆时偏移算法提出存储优化方案，更高效的利用GPU上各类存储器，以进一步降低数据访问所 造成的时间延迟。 |
| 关键词: 图形处理器；叠前逆时偏移；混合粒度；并行计算；存储优化 |
|
中图分类号: TP391
文献标识码: A
|
|
| The Hybrid Granularity Data Segmentation and Storage Optimization of Prestack Reverse-Time Migration Based on GPU |
|
HAN Fei,LI Wei1,2
|
1.( 1.Lenovo Beijing Co.LTD., Beijing 100094, China;2.
2.Lenovo Beijing Information Technology Co.LTD., Beijing 100094, China)
|
| Abstract: To improve the computational efficiency of prestack reverse-time migration,this paper adopts the MPI + CUDA parallel model to divide seismic data and parallel tasks.The MPI + CUDA parallel programming model is summarized and the hybrid granularity parallel algorithm of prestack reverse-time migration is proposed.Based on the special storage model of CUDA,we propose the storage optimization scheme for prestack reverse-time migration algorithm so as to reduce time delay caused by the data access with higher involvements of all kinds of memories on GPU. |
| Keywords: GPU;prestack RTM;hybrid granularity;parallel computing;storage optimization |
