資源描述:
《基于GPU的流動(dòng)影響枝晶生長(zhǎng)相場(chǎng)方法.pdf》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在行業(yè)資料-天天文庫(kù)。
1、第41卷第3期蘭州理工大學(xué)學(xué)報(bào)Vo1.41No.32015年6月Journalof.LanzhouUnivers..ity0fTechnologyJu軋2015文章編號(hào):1673—5196(2015)03—0019—07.基于GPU的流動(dòng)影響枝晶生長(zhǎng)相場(chǎng)方法朱昶勝一,賈金芳,馮力,肖榮振,王永賢(1.蘭州理工大學(xué)計(jì)算機(jī)與通信學(xué)院,甘肅蘭州730050;2.蘭州理工大學(xué)省部共建有色金屬先進(jìn)加工與再利用國(guó)家重點(diǎn)實(shí)驗(yàn)室,甘肅蘭州730050)摘要:將自適應(yīng)壓力迭代法修正的Sola算法與相場(chǎng)模型相結(jié)合,建立過冷熔體在強(qiáng)迫流動(dòng)狀態(tài)下枝
2、晶生長(zhǎng)的Sola-相場(chǎng)模型.針對(duì)傳統(tǒng)方法求解多場(chǎng)耦合相場(chǎng)模型時(shí)存在的計(jì)算量大,計(jì)算時(shí)間長(zhǎng),計(jì)算效率低等問題,提出基于CUDA+GPU軟硬件體系結(jié)構(gòu)的高性能計(jì)算方法.以高純丁二腈(SCN)過冷熔體為例,在CPU+GPU異構(gòu)平臺(tái)上實(shí)現(xiàn)了存在流動(dòng)時(shí)凝固微觀組織演化過程的并行求解,并對(duì)基于CPU+GPU平臺(tái)與CPU平臺(tái)的計(jì)算結(jié)果及計(jì)算效率進(jìn)行比較.結(jié)果表明,當(dāng)計(jì)算規(guī)模達(dá)到百萬量級(jí)時(shí),與CPU平臺(tái)上的串行算法相比,在CPU+GPU異構(gòu)平臺(tái)上達(dá)到了24.39倍的加速比,大大提高計(jì)算效率,并得到與串行計(jì)算相一致的結(jié)果.關(guān)鍵詞:相場(chǎng)方法;枝
3、晶生長(zhǎng);流動(dòng);GPU中圖分類號(hào):TG244;0246文獻(xiàn)標(biāo)識(shí)碼:AGPU-basedphase-fieldmethodfordendritegrowthwithflowaffection#ZHUChang—sheng·,JIAJin—fang,F(xiàn)ENGLi,XIAORong-zhen,W_ANGYong-xian(1.CollegeofComputerandCommunication,LanzhouUniv.ofTeeh,Lanzhou730050,China;2.StateKeyLaboratoryofAdvancedPr
4、o—eessingandRecyclingofNonferrousMetals,LanzhouUniv.ofTech.,Lanzhou730050,China)Abstract:Combiningthephase-fieldmodelwiththeSolaalgorithm,whichiscorrectedbyadaptiveitera—tionscheme。aSola-phasefieldmodelfordendritegrowthinundercooledmeltinthestateoftheforcedflowises
5、tablished.Aimedattheproblemsintraditiona1solutionofmulti-fieldcouplingsuchaslargecal—culationamount,longcomputationtime,andlowcomputationalefficiency,ahigh-performancecomputingmethodisproposedbasedODCUDA+GPUsoftwareandhardwarearchitecture.Takingthesuccinonitrite(SC
6、N)withhighpurityasanexample,theparalle1solutionstotheevolutionprocessofthesolidificationmicrostructureisimplementedonCPU+GPUheterogeneousplatformwhentheflowisavailable.ThecomputationalresultsandefficiencybasedontheCPU+GPUplatformandCPUplatformarecompared.Theresults
7、showthat,comparedwiththeserialalgorithmontheCPUplatform,aspeedupof24.39isa—chievedontheCPU+GPUheterogeneousplatformwhenthecalculationscaleisofmillionmagnitude,thecomputationa1efficiencyisgreatlyimproved。andtheparallelcomputationresultisidenticaltothatofseri—alcompu
8、tation.Keywords:phase-fieldmethod;dendritegrowth;flow;GPU從宏觀觀點(diǎn)看晶體的生長(zhǎng)過程實(shí)際上是一個(gè)熱有溫度差和濃度差引起的自然對(duì)流,這種對(duì)流在熔量、質(zhì)量和動(dòng)量等的輸運(yùn)過程.晶體生長(zhǎng)輸運(yùn)過程對(duì)體的凝固過程中是始終存在的,它對(duì)熔體凝固后的其