...
首页> 外文期刊>PASJ: Publications of the Astronomical Society of Japan >Hierarchical tree algorithm for collisional N-body simulations on GRAPE
【24h】

Hierarchical tree algorithm for collisional N-body simulations on GRAPE

机译:GRAPE碰撞N体模拟的层次树算法

获取原文
           

摘要

Wepresentanimplementationofthehierarchicaltreealgorithmontheindividualtimestepalgorithm(theHermitescheme)forcollisionalN-bodysimulations,runningontheGRAPE-9system,aspecial-purposehardwareacceleratorforgravitationalmany-bodysimulations.SuchacombinationofthetreealgorithmandtheindividualtimestepalgorithmwasnoteasyonthepreviousGRAPEsystemmainlybecauseitsmemoryaddressingschemewaslimitedonlytosequentialaccesstoafullsetofparticledata.ThepresentGRAPE-9systemhasanindirectmemoryaddressingunitandaparticlememorylargeenoughtostorealltheparticledataandalsothetreenodedata.Theindirectmemoryaddressingunitstoresinteractionlistsforthetreealgorithm,whichisconstructedonthehostcomputer,and,accordingtotheinteractionlists,forcepipelinescalculateonlytheinteractionsnecessary.Inourimplementation,theinteractioncalculationsaresignificantlyreducedcomparedtodirectN2summationintheoriginalHermitescheme.Forexample,wecanachieveaboutafactor30ofspeedup(equivalenttoabout17teraflops)againsttheHermiteschemeforasimulationofanN=106system,usinghardwareofapeakspeedof0.6teraflopsfortheHermitescheme.
机译:Wepresentanimplementationofthehierarchicaltreealgorithmontheindividualtimestepalgorithm(theHermitescheme)forcollisionalN-bodysimulations,runningontheGRAPE-9system,专机-purposehardwareacceleratorforgravitationalmany-bodysimulations.SuchacombinationofthetreealgorithmandtheindividualtimestepalgorithmwasnoteasyonthepreviousGRAPEsystemmainlybecauseitsmemoryaddressingschemewaslimitedonlytosequentialaccesstoafullsetofparticledata.ThepresentGRAPE-9systemhasanindirectmemoryaddressingunitandaparticlememorylargeenoughtostorealltheparticledataandalsothetreenodedata.Theindirectmemoryaddressingunitstoresinteractionlistsforthetreealgorithm,whichisconstructedonthehostcomputer,并且,accordingtotheinteractionlists,forcepipelinescalculateonlytheinteractionsnecessary.Inourimplementation,theinteractioncalculationsaresignificantlyreducedcomparedtodirectN2summationintheoriginalHermitescheme.Forexample,wecanachieveaboutafactor30ofspeedup(equivalenttoabout17teraflops)againsttheHermiteschemeforasimulationofan N = 106系统,使用最高速度为0.6 teraflops的硬件用于Hermites方案。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号