欢迎来到天天文库
浏览记录
ID:43484910
大小:217.02 KB
页数:6页
时间:2019-10-07
《FPGA流水线设计》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库。
1、DesignofVeryDeepPipelinedMultipliersforFPGAsAlexPanato,SandroSilva,FlávioWagner,MarceloJohann,RicardoReis,SergioBampiUniversidadeFederaldoRioGrandedoSul-InstitutodeInformáticaAvBentoGonçalves,9500,BlocoIV,PortoAlegre,RS,Brazile-mail:2、,johann,reis,bampi>@inf.ufrgs.brAbstractInthisworkweinvestigateamethodologytodesignthedeepestpipelinedcircuitsinFPGAs,startingfromVHDL.FPGAdeviceshavesomespecificcharacteristicsthatThisworkinvestigatestheuseofverydeeppipelinesforallowthedesignertoimplemen3、ta"gatelevel"pipelineimplementingcircuitsinFPGAs,whereeachpipelinewithoptimalperformance,theonlyremarkbeingthatthestageislimitedtoasingleFPGAlogicelement(LE).Thewordgateheremeansany4-inputfunctionwithasinglearchitectureandVHDLdesignofaparameterizedinteger4、output.LongerstageswillpresenttwicethedelayoflogicarraymultiplierispresentedandalsoanIEEE754elementsandwilluseanoutsideconnection.Shortercompliant32-bitfloating-pointmultiplier.WeshowhowtostagesdonottakeadvantageofthefactthattheFPGAcellwriteVHDLcellsthati5、mplementsuchapproach,andhowcanimplementanyfunctionwiththesamedelay.Theideathearraymultiplierarchitecturewasadapted.SynthesisalreadyappearsinanAlteraApplicationBrief[3],butweandsimulationwereperformedforAlteraApex20KEdidnotfinddescriptionsandresultsofameth6、odologyordevices,althoughtheVHDLcodeshouldbeportabletoimplementationanywhereelse.otherdevices.Forthisfamily,a16bitintegermultiplierDespitethefactthatFPGAarchitecturesdifferfromachievesafrequencyof266MHz,whilethefloatingpointvendortovendor,theystillpresent7、asetofbasiccommonunitreaches235MHz,performing235MFLOPSinanfeaturesthatallowbuildinggatelevelpipelinesinVHDL.FPGA.Additionalcellsareinsertedtosynchronizedata,Bydoingso,itispossibletoreuseandmapthedesigntowhatimposessignificantareapenalties.Thisandothermany8、differentdevices,andreuseitatthepressofaconsiderationstoapplythetechniqueinrealdesignsarebutton,incontrasttothefullcustomapproachalsoaddressed.Asacasestudy,wedevelopedanarchitectureforintegermultiplicationthatexploi
2、,johann,reis,bampi>@inf.ufrgs.brAbstractInthisworkweinvestigateamethodologytodesignthedeepestpipelinedcircuitsinFPGAs,startingfromVHDL.FPGAdeviceshavesomespecificcharacteristicsthatThisworkinvestigatestheuseofverydeeppipelinesforallowthedesignertoimplemen
3、ta"gatelevel"pipelineimplementingcircuitsinFPGAs,whereeachpipelinewithoptimalperformance,theonlyremarkbeingthatthestageislimitedtoasingleFPGAlogicelement(LE).Thewordgateheremeansany4-inputfunctionwithasinglearchitectureandVHDLdesignofaparameterizedinteger
4、output.LongerstageswillpresenttwicethedelayoflogicarraymultiplierispresentedandalsoanIEEE754elementsandwilluseanoutsideconnection.Shortercompliant32-bitfloating-pointmultiplier.WeshowhowtostagesdonottakeadvantageofthefactthattheFPGAcellwriteVHDLcellsthati
5、mplementsuchapproach,andhowcanimplementanyfunctionwiththesamedelay.Theideathearraymultiplierarchitecturewasadapted.SynthesisalreadyappearsinanAlteraApplicationBrief[3],butweandsimulationwereperformedforAlteraApex20KEdidnotfinddescriptionsandresultsofameth
6、odologyordevices,althoughtheVHDLcodeshouldbeportabletoimplementationanywhereelse.otherdevices.Forthisfamily,a16bitintegermultiplierDespitethefactthatFPGAarchitecturesdifferfromachievesafrequencyof266MHz,whilethefloatingpointvendortovendor,theystillpresent
7、asetofbasiccommonunitreaches235MHz,performing235MFLOPSinanfeaturesthatallowbuildinggatelevelpipelinesinVHDL.FPGA.Additionalcellsareinsertedtosynchronizedata,Bydoingso,itispossibletoreuseandmapthedesigntowhatimposessignificantareapenalties.Thisandothermany
8、differentdevices,andreuseitatthepressofaconsiderationstoapplythetechniqueinrealdesignsarebutton,incontrasttothefullcustomapproachalsoaddressed.Asacasestudy,wedevelopedanarchitectureforintegermultiplicationthatexploi
此文档下载收益归作者所有