资源描述：
《A Low-Power Vector Processor Using Logarithmic Arithmetic for Handheld 3D Graphics Systems》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、ALow-PowerVectorProcessorUsingLogarithmicArithmeticforHandheld3DGraphicsSystemsByeong-GyuNamandHoi-JunYooDept.ofEECS,KoreaAdvancedInstituteofScienceandTechnology(KAIST)373-1,Guseong-dong,Yuseong-gu,Daejeon,305-701,RepublicofKoreaAbstract-Alow-power,high-performa
2、nce4-way32-bitvectorA.NumberSystemprocessorisdevelopedforhandheld3Dgraphicssystems.ItTheproposedarithmeticunitisbasedonthehybridcontainsafloating-pointunifiedmatrix,vector,andelementaryapproachoftheFLPandthelogarithmicnumbersystemfunctionunit.Byutilizingthelogar
3、ithmicarithmetic,theunit(LNS)introducedin[4],whereoperationsarereducedintoachievessingle-cyclethroughputforalltheseoperationsexceptsimpleronesintheLNSwhiletheadditionandsubtractionareforthematrix-vectormultiplicationwith2-cyclethroughput.TheperformedinFLPsinceth
4、eLNSadditionandsubtractionprocessorfeaturedbythisfunctionunit,cascadedinteger-floatdatapath,reconfigurationofdatapath,operandforwardinginrequirenonlineartermevaluations.Thelogarithmicandlogarithmicdomain,andvertexcachetakes9.7mm2in0.18µmantilogarithmicconverters
5、betweentheFLPandtheLNSareCMOStechnologyandachieves141Mvertices/sforgeometryproposedforthishybridnumbersystem(HNS).transformationand12.1Mvertices/sforOpenGLtransformationxXandlightingat200MHzwith86.6mWpowerconsumption.15-entryLUT(64B)8entryLUT(56B)I.Icidieibicidi
6、biNTRODUCTIONThehandheldgraphicsprocessingunits(GPUs)incorporatemm>>cim>>dim>>eiff>>cif>>divectorprocessors,knownasshaders,intheir3DgraphicsCSACSApipelinestagestoprovidemorerealisticimages[1].In[2],aCSAvertexshaderisproposedwith16-wayfloating-point(FLP)CSAmultip
7、liersforthefastgeometrytransformationanditCSACPAconsumedalargesiliconareaandpower.Forthepower-andCPAarea-efficientdesignoftheshaders,amultifunctionunitwaseaif+biproposedin[3].However,itwasafixed-pointunitanddidn’teaim+bidealwiththematrix-vectormultiplication,fre
8、quentlyusedfor(a)Logarithmicconverter(b)Antilogarithmicconverter3Dgeometrytransformations.Fig.1.ProposednumberconvertersInthispaper,a4-way32-bitFLPvectorprocessoris1)
