简介:ThispaperproposesanovelstrategybasedonfragmentmeshesofShiueetal.forGPUrenderingofcompositesubdivisionsurfaces.Twoenumerationsystemsareestablishedtolabeltheprimitivesofeachfragmentmesh.Asector-layer-indexenumerationsystemisresponsibleforretrievingproximitiesforsubdivisionmaskswhileasector-indexenumerationsystemdesignatesa2DtexturebufferinGPU.Recurringtothefreeconversionbetweenthem,ourapproachmaygetridoflookuptablesthataredesignedtorecordsubdivisionmasks.Inaddition,relativelysmallcompositesubdivisionmasksmakeiteasytodevelopautomaticallyretrievingtechniques.Finally,ascenterverticesareoftenirregular,theircomputationisrelatedtoanaveragewithalterablenumberofitems.ConsideringthatvariableloopisnotefficientinGPU,weevaluatethecentervertexofeachfragmentmeshusingthelinearcombinationofitslevel0,level1andlimitpositionsinsteadofaveragingschemes.ExperimentsdemonstratethatourapproachgenerallyoutperformsthatofShiueetal.inFPSbyalongway.
简介:Directisosurfacevolumerenderingisthemostprominentmodernmethodformedicaldatavisualization.Itisbasedonfindingintersectionpointsbetweentherayscorrespondingtopixelsonthescreenandisosurface.Thisarticledescribesatwo-passalgorithmforacceleratingthemethodonthegraphicprocessingunit(GPU).Onthefirstpass,theintersectionswiththeisosurfacearefoundonlyforasmallnumberofrays,whichisdonebyrenderingintoalower-resolutiontexture.Onthesecondpass,theobtainedinformationisusedtoefficientlycalculatetheintersectionpointsofalltheother.Thenumberofraystouseduringthefirstpassisdeterminedbyusinganadaptivealgorithm,whichrunsonthecentralprocessingunit(CPU)inparallelwiththesecondpassoftherendering.Theproposedapproachallowstosignificantlyspeedupisosurfacevisualizationwithoutqualityloss.Experimentsshowaccelerationupto10timesincomparisonwithacommonraycastingmethodimplementedonGPU.Totheauthors’knowledge,thisisthefastestapproachforraycastingwhichdoesnotrequireanypreprocessingandcouldberunoncommonGPUs.
简介:摘要就大部分学科而言,图数据是其基础理论,尤其在数学领域及计算机领域,通过怎样的途径可以使图算法提升其计算效率是当前研究的一个重点内容,而现今算法已日趋成熟,传统形式的图算法早已不能与之相适应,因而,并行图算法的相关研究受到越来越多人的关注。而相较传统形式的CPU,GPU的运算能力更加强大,因而受到的关注度也越来越高,使得图算法在该领域也得到了有效的发展。本文主要简析了现阶段图论算法中引入的GPU领域中的加速技术。
简介:AnewGraphicsProcessingUnit(GPU)parallelizationstrategyisproposedtoacceleratesparsefiniteelementcomputationforthreedimensionalelectromagneticanalysis.TheparallelizationstrategyisemployedbasedonanewcompressionformatcalledslicedELLFour(slicedELL-F).TheslicedELL-Fformat-basedparallelizationstrategyisdesignedforhasteningmanyaddition,dotproduct,andSparseMatrixVectorProduct(SMVP)operationsintheConjugateGradientNorm(CGN)calculationoffiniteelementequations.ThenewimplementationofSMVPonGPUsisevaluated.TheproposedstrategyexecutedonaGPUcanefficientlysolvesparsefiniteelementequations,espe-ciallywhentheequationsarehugesparse(sizeofmostrowsinacoefficientmatrixislessthan8).NumericalresultsshowtheslicedELL-Fformat-basedparallelizationstrategycanreachsigni?cantspeedupscomparedtoCompressedSparseRow(CSR)format.