1.State Key Laboratory of Geomechanics and Geotechnical Engineering Safety, Institute of Rock and Soil Mechanics, Chinese Academy of Sciences, Wuhan 430071, China
2.School of Engineering Science, University of Chinese Academy of Science, Beijing 100049, China
3.School of Resources and Environmental Engineering, Wuhan University of Technology, Wuhan 430070, China
4.School of Highway, Chang'an University, Xi'an 710064, China
In recent years, the material point method (MPM) has become an important large-deformation numerical simulation method in geotechnical engineering and is widely utilized to study issues such as landslides, foundations, dam failures, and water-soil gushing in shield tunnels. As the scale and complexity of application scenarios increase, the accuracy requirements and efficiency needs for numerical methods also rise, resulting in higher computational costs that restrict the further application of MPM in large-scale geotechnical engineering problems. The simulation efficiency of MPM improves significantly by introducing parallel acceleration technology; however, specific issues, such as program architecture and extensibility, still limit their development in engineering applications. This study proposes an improved GPU acceleration strategy for MPM by incorporating modular programming concepts, efficient data structures, and data competition handling methods to establish a high-performance and easily extendable program architecture. The software's performance is evaluated by simulating aluminum rod collapse experiments and the slope failure process. The results indicated that the software achieves effective parallelization and demonstrates approximately a 10% improvement over the existing Taichi-GPU MPM. Finally, the proposed MPM method is applied to simulate the Xinmo landslide, and the computational efficiency improves by approximately 20 times when the number of material points increases by about 2.5 times.
在物质点法模拟中,由于物质点移动影响,网格信息的数据结构通常是稀疏的,所以对数据结构的处理会大幅度影响内存使用和计算效率。Hoetzlein等[18]在GPU上扩展计算机图形学中最流行的OpenVDB稀疏存储方案,提出了高效内存架构的GVDB点,以支持动态拓扑变化。Setaluri [19]和Gao [20]等分别证明SPGrid是(MPM)中高效的数据结构。物质点信息通常是存储在结构体数组(array of struct,AoS)或数组结构体(structure of array,SoA)中。顺序线程访问顺序内存地址时,SoA促进物质点数据的合并内存访问。然而,为了保持这种高效的数据访问模式,物质点需要在每个时间步之后重新排序,造成SoA在序列化的收集和分散操作中效率较低。相反,AoS更容易映射到物质点的概念,并且由于单个粒子数据的局域性,在非合并内存访问模式的情况下表现良好。然而,这样的内存布局阻止了粒子数据的合并读写,从而在可能合并时显著抑制GPU和向量化CPU的性能。为综合SoA和AoS的优缺点,Wang等[21]以MPM为中心提出了数组‒结构‒数组的数据结构,使用AoSoA分层方式存储粒子数据,粒子重新组织在低级桶和高级桶中以保存内存访问和数据传输的效率。
HarlowF H.The particle-in-cell computing method for fluid dynamics[J].Methods in Computational Physics,1964,3: 319‒343.
[2]
SulskyD, ChenZ, SchreyerH L.A particle method for history-dependent materials[J].Computer methods in applied mechanics and engineering,1994,118(1/2):179‒196. doi:10.1016/0045-7825(94)90112-0
[3]
AndersenS, AndersenL.Modelling of landslides with the material-point method[J].Computational Geosciences,2010, 14:137‒147. doi:10.1007/s10596-009-9137-y
[4]
ZhongZuliang, HeKaiyuan, SongYixiang,et al.Large-displacement landslides based on affine velocity matrix-improved material point method[J].Chinese Journal of Geotechnical Engineering,2022,44(9):1626‒1634.
XieYanfang, LiXinpo, ZhaoShux,et al.MPM-based numerical analysis of the kinematic characteristics of Xinmo landslide in Maoxian County,Sichuan,China[J].Mountain Research,2018,36(4):589‒597.
WangBin, FengXiating, PanPengzhi,et al.Slope failure analysis using the material point method[J].Chinese Journal of Rock Mechanics and Engineering,2017,36(9):2146‒2155.
LiangDongfang, ZhaoXuanyu, KenichiSoga.Simulation of overtopping and seepage induced dike failure using two-point MPM[J].Soils and Foundations,2020,60(4):978‒988. doi:10.1016/j.sandf.2020.06.004
[12]
XieXiaochuang, FrancescaCeccato, ZhouMingliang,et al.Hydro-mechanical behaviour of soils during water-soil gushing in shield tunnels using MPM[J].Computers and Geotechnics,2022,145:104688. doi:10.1016/j.compgeo.2022.104688
[13]
SunZheng.A study for the improved B-spline material point method[D].Zhejiang:Zhejiang University,2018.
[14]
孙政.B样条物质点法的算法改进研究[D].浙江:浙江大学, 2018.
[15]
HuYuanming, FangYu, GeZiheng,et al.A moving least squares material point method with displacement discontinuity and two-way rigid body coupling[J].ACM Transactions on Graphics(TOG),2018,37(4):1‒14. doi:10.1145/3197517.3201293
[16]
YuanWeihai,Wang Haocheng Liu Kang,et al.Analysis of large deformation geotechnical problems using implicit generalized interpolation material point method[J].Journal of Zhejiang University‒SCIENCE A,2021,22(11):909‒923. doi:10.1631/jzus.a2100219
[17]
HuangPeng, ZhangXiong, MaShang,et al.Shared memory OpenMP parallelization of explicit MPM and its application to hypervelocity impact[J].Computer Modeling in Engineering & Sciences,2008,38(2):119‒148.
[18]
DongY, WangD, RandolphM F.A GPU parallel computing strategy for the material point method[J].Computers and Geotechnics,2015,66:31‒38. doi:10.1016/j.compgeo.2015.01.009
[19]
LeiChunxia, GanYong.Study of the thickness rffects on copper films heated by the femtosecond laser[J].Bulletin of Science and Technology,2018,34(8):6‒9.
XiaXilin, LiangQiuhua.A GPU-accelerated smoothed particle hydrodynamics(SPH) model for the shallow water equations[J].Environmental modelling & software,2016,75:28‒43. doi:10.1016/j.envsoft.2015.10.002
[22]
ZhangWei, ZhongZhihao, PengChong,et al.GPU-accelerated smoothed particle finite element method for large deformation analysis in geomechanics[J].Computers and Geotechnics,2021,129:103856. doi:10.1016/j.compgeo.2020.103856
[23]
HoetzleinR K.GVDB:Raytracing sparse voxel database structures on the GPU[C]//Proceedings of High Performance Graphics,New York:Association for Computing Machinery(ACM),2016:109-117.
[24]
SetaluriR, AanjaneyaM, BauerS,et al.SPGrid:A sparse paged grid structure applied to adaptive smoke simulation[J].ACM Transactions on Graphics(TOG),2014,33(6):1‒12. doi:10.1145/2661229.2661269
[25]
GaoMing, WangXinlei, WuKui,et al.GPU optimization of material point methods[J].ACM Transactions on Graphics (TOG),2018,37(6):1‒12. doi:10.1145/3272127.3275044
[26]
WangXinlei, QiuYuxing, Slattery StuartR,et al.A massively parallel and scalable multi-GPU material point method[J].ACM Transactions on Graphics(TOG),2020,39(4):30:1‒30. doi:10.1145/3386569.3392442
[27]
BuiH H, FukagawaR, SakoK,et al.Lagrangian meshfree particles method(SPH) for large deformation and failure flows of geomaterial using elastic-plastic soil constitutive model[J].International Journal for Numerical and Analytical Methods in Geomechanics,2008,32(12):1537‒1570. doi:10.1002/nag.688
[28]
LiJianguo, WangBin, JiangQuan,et al.Development of an adaptive CTM‒RPIM method for modeling large deformation problems in geotechnical engineering[J].Acta Geotechnica,2021:1‒19. doi:10.1007/s11440-021-01416-1
[29]
HuangPeng.Material point method for metal and soil impact dynamics problems[D].Beijing:Tsinghua University,2011.
[30]
黄鹏.金属及岩土冲击动力学问题的物质点法研究[D].北京:清华大学,2011.
[31]
HuYuanming, LiTzumao, LukeAnderson,et al.Taichi:A language for high-performance computation on spatially sparse data structures[J].ACM Transactions on Graphics(TOG),2019,38(6):1‒16. doi:10.1145/3355089.3356506
[32]
张雄,廉艳平,刘岩,等.物质点法[M].北京:清华大学出版社,2013.
[33]
ZhaoShuxi, HeSiming, LiXinpo,et al.The Xinmo rockslide-debris avalanche:An analysis based on the three-dimensional material point method[J].Engineering Geology,2021,287:106109. doi:10.1016/j.enggeo.2021.106109
[34]
FanXuanmei, XuQiang, GianvitoScaringi,et al.Failure mechanism and kinematics of the deadly June 24th 2017 Xinmo landslide,Maoxian,Sichuan,China[J].Landslides, 2017,14:2129‒2146. doi:10.1007/s10346-017-0907-7
[35]
GianvitoScaringi, FanXuanmei, XuQiang,et al.Some considerations on the use of numerical methods to simulate past landslides and possible new failures:the case of the recent Xinmo landslide(Sichuan,China)[J].Landslides,2018,15:1359‒1375. doi:10.1007/s10346-018-0953-9