Computer Architecture and Parallel Computing

Computer Architecture and Parallel Computing

ID:39065527

大小:560.13 KB

页数:65页

时间:2019-06-24

Computer Architecture and Parallel Computing_第1页
Computer Architecture and Parallel Computing_第2页
Computer Architecture and Parallel Computing_第3页
Computer Architecture and Parallel Computing_第4页
Computer Architecture and Parallel Computing_第5页
资源描述:

《Computer Architecture and Parallel Computing》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、ComputerArchitectureandParallelComputing并行结构与计算Lecture2-PipeliningPengLiuDept.ofInfo.Sci.&Elec.Engg.ZhejiangUniversityliupeng@zju.edu.cnMay18,2015•MicrocodingbecamelessattractiveasgapbetweenRAMandROMspeedsreduced•Complexinstructionsetsdifficulttopipeline,sodifficulttoincreaseperformanceasgatecou

2、ntgrew•IronLawexplainsarchitecturedesignspace–Tradeinstructions/program,cycles/instruction,andtime/cycle•Load-StoreRISCISAsdesignedforefficientpipelinedimplementations–Verysimilartoverticalmicrocode–InspiredbyearlierCraymachines(moreontheselater)2AnIdealPipelinestagestagestagestage1234•Allobject

3、sgothroughthesamestages•Nosharingofresourcesbetweenanytwostages•Propagationdelaythroughallpipelinestagesisequal•TheschedulingofanobjectenteringthepipelineisnotaffectedbytheobjectsinotherstagesTheseconditionsgenerallyholdforindustrialassemblylines,butinstructionsdependoneachother!3PipelinedMIPSTo

4、pipelineMIPS:•FirstbuildMIPSwithoutpipeliningwithCPI=1•Next,addpipelineregisterstoreducecycletimewhilemaintainingCPI=14UnpipelinedDatapathforMIPSPCSrcRegWriteMemWriteWBSrcbrrindjabspc+40x4AddAddclkweclkrs1rs2PCaddr31rd1weinstwsaddrALUwdrd2clkInst.GPRszrdataMemoryDataImmMemoryExtwdataALUControlOp

5、CodeRegDstExtSelOpSelBSrczero?5HardwiredControlTableOpcodeExtSelBSrcOpSelMemWRegWWBSrcRegDstPCSrcALU*RegFuncnoyesALUrdpc+4ALUisExtImmOpnoyesALUrtpc+416ALUiuuExtImmOpnoyesALUrtpc+416LWsExtImm+noyesMemrtpc+416SWsExt16Imm+yesno**pc+4BEQZz=0sExt16*0?nono**brBEQZz=1sExt16*0?nono**pc+4J***nono**jabsJA

6、L***noyesPCR31jabsJR***nono**rindJALR***noyesPCR31rindBSrc=Reg/ImmWBSrc=ALU/Mem/PCRegDst=rt/rd/R31PCSrc=pc+4/br/rind/jabs6PipelinedDatapath0x4Addwers1rs2PCaddrrd1werdataIRwsaddrwdrd2ALUGPRsrdataDataInst.MemoryImmMemoryExtwdatawritefetchdecode&Reg-fetchexecutememory-backphasephasephasephasephaseC

7、lockperiodcanbereducedbydividingtheexecutionofaninstructionintomultiplecyclest>max{t,t,t,t,t}(=tprobably)CIMRFALUDMRWDMHowever,CPIwillincreaseunlessinstructionsarepipelined7“IronLaw”ofProcessorPerformanceTime=InstructionsCyc

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。