8 Pages

lec10

Course: CS 422, Spring 2010
School: IIT Kanpur
Rating:
 
 
 
 
 

Word Count: 324

Document Preview

- LECTURE 10 Data Hazards CC1 CC2 CC3 CC4 CC5 CC6 CC7 CC8 CC9 CC10 CC11 CC12 LD F4, 0(R2) IF ID EX MEM WB IF ID STL M1 M2 M3 M4 M5 M6 M7 MEM MULTD F0, F4, F6 RAW hazards cause more stalls now CC1 CC2 CC3 CC4 CC5 MULTD F0, F4, F6 IF ... ... ADDD F2, F4, F6 ... LD F2, 0(R2) CC6 CC7 CC8 CC9 CC10 CC11 M6 M7 MEM WB MEM WB ID M1 M2 M3 M4 M5 IF ID EX MEM WB IF ID EX IF ID A1 A2 A3 A4 IF...

Register Now

Unformatted Document Excerpt

Coursehero >> India >> IIT Kanpur >> CS 422

Course Hero has millions of student submitted documents similar to the one
below including study guides, practice problems, reference materials, practice exams, textbook help and tutor support.

Course Hero has millions of student submitted documents similar to the one below including study guides, practice problems, reference materials, practice exams, textbook help and tutor support.
- LECTURE 10 Data Hazards CC1 CC2 CC3 CC4 CC5 CC6 CC7 CC8 CC9 CC10 CC11 CC12 LD F4, 0(R2) IF ID EX MEM WB IF ID STL M1 M2 M3 M4 M5 M6 M7 MEM MULTD F0, F4, F6 RAW hazards cause more stalls now CC1 CC2 CC3 CC4 CC5 MULTD F0, F4, F6 IF ... ... ADDD F2, F4, F6 ... LD F2, 0(R2) CC6 CC7 CC8 CC9 CC10 CC11 M6 M7 MEM WB MEM WB ID M1 M2 M3 M4 M5 IF ID EX MEM WB IF ID EX IF ID A1 A2 A3 A4 IF ID EX MEM WB IF ID EX MEM MEM WB WAW hazard: an example WB Handling WAW Hazards Occurs only when the result of ADDD is overwritten without any instruction using it! Otherwise, RAW hazard stall would have occurred Hazard can be detected in ID stage of latter instruction Two ways to handle: Delay issue of load until ADDD enters MEM Stamp out result of ADDD Control Hazard Complications An example: DIVF F0, F2, F4 // Finishes last; excepn. ADDF F10, F10, F8 // Finishes first SUBF F12, F12, F14 // Finishes second Out-of-order completion causes problems! Precise are exceptions difficult to implement Achieving Precise Exceptions Approach 1: Ostrich algorithm Don't care May be provide a slower precise mode Example: special instructions to check for FP exceptions Approach 2: allow instruction issue to continue only if previous instructions will complete without exception Stall to maintain precise exceptions Achieving Precise Exceptions (continued) Approach 3: save state to undo Two possibilities History file: keep track of original value of registers Future file: keep track of current value; main register file updated after all previous instructions are done More buffer space required Hazard checks and control become very complex Achieving Precise Exceptions (continued) Approach 4: imprecise, but keep enough state for OS to recover Keep track of incomplete instructions OS then runs those instructions before returning control Complicated to execute these instructions properly! Further Topics... Instruction Level Parallelism (ILP)
Find millions of documents on Course Hero - Study Guides, Lecture Notes, Reference Materials, Practice Exams and more. Course Hero has millions of course specific materials providing students with the best way to expand their education.

Below is a small sample set of documents:

IIT Kanpur - CS - 422
LECTURE - 11Instruction Level ParallelismPipelining achieves Instruction LevelParallelism (ILP)Multiple instructions in parallelBut, problems with pipeline hazardsCPI = Ideal CPI + stalls/instructionStalls = Structural + Data (RAW/WAW/WAR) +Contro
IIT Kanpur - CS - 422
LECTURE - 12ILP: RecallImproving ILP = reducing stallsLoop unrolling enlarges the basic blockMore parallelismMore opportunity for better schedulingDependences:Data dependenceName dependenceControl dependenceHandling Control DependenceControl de
IIT Kanpur - CS - 422
LECTURE - 13Dynamic SchedulingBetter than static schedulingScoreboarding:Used by the CDC 6600Useful only within basic blockWAW and WAR stallsTomasulo algorithm:Used in IBM 360/91 for the FP unitMain additional feature: register renaming toavoid
IIT Kanpur - CS - 422
LECTURE - 14Dealing with Control HazardsSoftware techniques:Branch delay slotsSoftware branch predictionCanceling or nullifying branchesMisprediction rates can be highWorse if multiple issue per cycleHence, hardware/dynamic branch predictionBranc
IIT Kanpur - CS - 422
LECTURE - 15Further Topics in ILPMultiple issueSoftware supportHardware supportIncreasing ILP throughMultiple IssueWith at most one issue per cycle, min CPIpossible is 1But there are multiple functional unitsHence use multiple issueTwo ways to
IIT Kanpur - CS - 422
LECTURE - 16Hardware Support forSpeculationConditional or predicated instructionsExecute on condition, annul otherwiseExample: conditional moveif (A = 0) cfw_ S = T; BNEZ R1, LMOV R2, R3CMOVZ R2, R3, R1L:Control dependence has been eliminatedD
IIT Kanpur - CS - 422
LECTURE - 17LECTURE - 17SpeculationWish to move instructions across branchesTo eliminate possible stallsFor better schedulingAppropriate conditional instructions may notalways existExample:if (N = 0) cfw_A = *X; else cfw_A+;Speculation: An E
IIT Kanpur - CS - 422
Memory HierarchyTwo principles:Smaller is fasterPrinciple of localityProcessor speed grows much faster thanmemory speedRegisters Cac he Memory DiskUpper level vs. lower levelCache designCache Design QuestionsCache is arranged in terms of blocks
IIT Kanpur - CS - 422
LECTURE - 19Topics for TodayCache PerformanceCache Misses: The Three C'sImproving the Cache Miss RateScribe?Cache PerformanceMiss rate is an important metricBut not the only oneAvg. mem. access timeHit time Miss rate Miss penaltyHit time, Miss
IIT Kanpur - CS - 422
LECTURE - 20LECTURE - 20Topic for TodayReducing Cache Miss PenaltyScribe?Technique-1: Prioritize ReadMisses over WritesWrite-through cache => write-bufferBeware of consistencyExample: store x, load y, load x and y in thexsame blockPossible sol
IIT Kanpur - CS - 422
LECTURE - 21Topics for TodayHit-time reduction techniquesVirtual memoryScribe for today?Small and Simple CachesKeep the cache smallFasterCan fit inside processorTrade-off: tags within processor, data outsideKeep the cache simpleDirect-mapped =>
IIT Kanpur - CS - 422
LECTURE - 22Topics for TodayMain memoryScribe for today?Main MemoryDRAM versus SRAMDRAM is cheaper, but slowerReducing the number of pinsAt the cost of some performanceAddress = RAS + CASPerformance metrics: latency and bandwidth#cycles to send
IIT Kanpur - CS - 422
LECTURE - 23Topics RemainingHW2 handed out today/tomorrowMultiprocessors: 4 lecturesInter-connection networks: 1 lectureI/O: 2 lecturesReview: 1 or 2 lecturesTake-home part of end-sem handed outSpecial topics:Vector processorsPower optimization
IIT Kanpur - CS - 422
LECTURE - 24Topic for Today's LectureMultiprocessingParallel applicationsCache coherenceScribe for today?Multiprocessing: ClassificationMultiprocessingSIMDMIMDCentralizedshared memoryPhysicallydistributed memoryDistributed sharedmemory (DSM
IIT Kanpur - CS - 422
LECTURE - 25Topic for Today's LectureCache coherence protocolsScribe for today?Cache CoherenceIn what kind of multi-processors do we needcache coherence?What are the kinds of data which arecached?Shared (read) data repl icationPrivate data mig
IIT Kanpur - CS - 422
LECTURE - 26Lecture OutlineCache coherence protocols (continued)Snooping-based schemesDirectory-based schemesScribe for today?Snooping-Based ProtocolsApplicable for write-through as well as writeback cachesOptimizations:Shared/Exclusive bitWrite
IIT Kanpur - CS - 422
LECTURE - 27Lecture OutlineSynchronization mechanisms, andconsistency modelsScribe for today?SynchronizationRequired since communication is throughshared memorySynchronization primitivesInvolve atomic read-and-write of a memorylocationAtomic ex
IIT Kanpur - CS - 422
LECTURE - 28Lecture OutlineConsistency modelsScribe for today?Sequential ConsistencySequential consistency: result of executionsame as if:Accesses executed by a processor are in orderAccesses among different processors areinterleavedThat is, the
IIT Kanpur - CS - 422
LECTURE - 29Lecture OutlineInterconnection networksIBM's Blue GeneI/O and BusesScribe for today?Interconnection NetworksNetworks at three levels:Massively Parallel Processor (MPP) Network,within about 25m maxLAN: within about a few km maxWAN: l
IIT Kanpur - CS - 422
LECTURE - 30Lecture OutlineI/O and BusesSome queuing theoryScribe for today?Buses for CommunicationBetween CPU/Memory, and with I/O devicesAdvantages:Low-costFlexible/versatileDisadvantage:Communication bottleneckBandwidth limited due to lengt
IIT Kanpur - CS - 422
LECTURE - 31Lecture OutlineSome queuing theoryUNIX Fast File SystemScribe for today?I/O PerformanceProducer-Server ModelThroughput vs. Response TimeResponse time and think timeQueuing theoryArrival rate, service time, utilizationLittle's lawSq
IIT Kanpur - CS - 422
LECTURE - 32Lecture OutlineLog-Structured File System (LFS) [RO91]RAIDScribe for today?Log-Structured File SystemTechnological under-pinnings:Disk I/O becoming bottleneck since CPUs aregetting fasterDisk I/O dominated by writes, since reads mostl
IIT Kanpur - CS - 422
LECTURE - 33Lecture OutlineVector ProcessorsScribe for today?Why Vector ProcessingDeep pipeline => more parallelismBut more dependencesNeed to fetch and issue many instructions (Flynnbottleneck)Same issues with multiple-issue processorOperations
IIT Kanpur - CS - 422
LECTURE - 34Key Take-Away IdeasQuantitative approach to designAmdahl's lawDesign to match technology trendInterface designPipelining, non-uniformity is badGolden rule: preserve programmer's viewComplexity in hardware vs. softwareKey Take-Away Ide
University of the West Indies at Mona - MSB - 6040
Contents1.1.Background. 21.2.The Research Problem . 21.3.The Research Questions . 31.4.Research Method . 31.5.Design Type . 31.6.Desk Research. 41.7.Collection of Secondary Data. 52.1.Qualitative Research. 52.2.Sample D
Unicamp - PQI - 2110
PQI2110 QUMICATECNOLGICA GERALINTRODUO SLIGAES QUMICAS1Tabela Peridica12345678910111213141516171823HHeHydrogen1HeliumLiBeBCNOFNeLithiumBerylliumBoronCarbonNitrogenOxygenFluorineNeonNaMgAlSiPSClArAlumi
Ill. Chicago - IDS - 355
MIDTERM 2CHAPTER 8SThe Transportation Model Transportation Problem Involves finding the lowest-cost plan for distributing stocks of goods or suppliesfrom multiple origins to multiple destinations that demand the goods. Transportation Model: Applicat
Ill. Chicago - IDS - 355
CHAPTER 1: introduction to operations managementWhat is operations? The part of a business organization that is responsible for producing goods or services The management of systems or processes that create goods and/or provide servicesSupply & Demand
Ill. Chicago - IDS - 355
IDS 355IDS 355Summer 2011Assignment 2Due Friday 7/1/201140 pointsInstructionsDeadlineThis assignment may be submitted anytime until Friday, July 1st at 11:55 p.m. Because solutions will beposted soon after the deadline, late submissions may not b
Ill. Chicago - IDS - 355
IDS 355IDS 355Summer 2011Assignment 1Due MONDAY JUNE 27 201140 pointsInstructionsDeadlineThis assignment may be submitted anytime until Monday, June 27th at 11:55 p.m. Because solutions willbe posted soon after the deadline, late submissions may
Ill. Chicago - IDS - 355
IDS 355Introduction to Operations ManagementSummer 2011INSTRUCTOR:Office:Phone:Email:Prof. Aris OukselUH 2411312-996-0771aris@uic.eduClass Time:Days:Classroom:04:15pm-06:45pmTuesday, ThursdayLecture Center F1Office Hours:By appointment o
University of Phoenix - AED - 222
AED 222Planning PyramidAxia College7/14/11Bloom's Taxonomy within a Planning PyramidTeachers today use several different factors of the planning pyramid to help ensurestudents understand their work. Many teachers find it extremely helpful to write d
Lake County - SPEECH - 121
Chapter 1 Speaking In PublicPublic Speaking vs. Conversation Similaritieso Oraganizing thoughts logicallyo Tailoring message to audienceo Telling story for maximum impacto Adapting to feedback Differenceso Public speaking is more highly structured
Lake County - SPEECH - 121
Chapter 2 Ethics and Public SpeakingEthics deal with issues of r ight and wrong in human affairs.Ethical Speaking Guidelines Make ethically sound goals Be fully prepared Be honest Avoid name calling and abusive language Put priciples in practicePl
Lake County - SPEECH - 121
Chapter 3 L isteningHearing vibration of sound waves on car drumsL istening understanding what is being saidTypes of Listening Appreciative listening for pleasure/enjoyment Empathic listening to provide emotional support for thespeaker Comprehensiv
Lake County - SPEECH - 121
Chapter 4 Selecting a Topic and PurposeBrainstorming generating ideas by the free association of wordsGeneral purpose broad goal of speech. Either to inform or to persuade.Specific purpose single infinitive phrase that states what the speaker hopesto
Lake County - SPEECH - 121
Chapter 6 Gathering Materials and In terviewingGathering M ater ialsUsing YourOwn Knowledge and ExperienceL ibrariansCataloguePeriodical databasesNewspapersReferences bookso Encyclopediaso Yearbookso Quotation bookso Biographical aids Search
Lake County - SPEECH - 121
Chapter 8 Organizing the SpeechOrganization is ImportantSpecific Purpose: To inform my audience about major uses ofhypnosis.Central Idea:The major uses of hypnosis today are tocontrol pain insurgery, to help people stopsmoking, and to helps tuden
Lake County - SPEECH - 121
Chapter 9 Beginning and Ending the SpeechThe IntroductionGet Attention and Interest Relate the topic to the audience Tate the importance of your topic Startle the audience Arouse the curiosity of the audience Question the audience Begin with a que
Lake County - SPEECH - 121
Chapter 10 Outlining the SpeechPreparation Outline a detailed outline developed during the process of speechpreparation that includes the title, specific purpose, central idea, introduction, mainpoints, sub-points, connections, conclusion, and bibliogr
Lake County - SPEECH - 121
Chapter 13 Using Visual AidsKinds of Visual Aids Objects Models Photographs Drawings Graphs Charts Transparencies Video/multimedia presentation The speaker himselfGuidelines for Preparing Visual Aids Prepare visual aids in advance Keep visual
Lake County - SPEECH - 121
Chapter 14 Speaking to InformTypes of Informative Speeches Speeches about objects Speeches about processes Speeches about events Speeches about conceptsGuidelines for Informative Speaking Dont over estimate what the audience knows Relate the subje
Lake County - SPEECH - 121
1Rahul Saha6/21/2011CMM 121Informative Speech: TornadoesPurpose: To inform the audience about tornadoesThesis: Tornados are dangerous natural disasters and should not be taken lightly. Ibelieve that knowledge of tornadoes can help us better survive
Lake County - SPEECH - 121
Tornados!RahulSahaClicktoeditMastersubtitlestyle7/15/11WhatIsaTornado?http:/www.youtube.com/watch?v=r4iofIQTX7o&feature=related7/15/11Cincinnati,OH7/15/11AccordingtotheAmericanRedCrosstornadoesoccurwhenthunderstormsdevelopinwarm,moistairinadv
Lake County - SPEECH - 121
Lake County - SPEECH - 121
Lake County - SPEECH - 121
Rahul Saha7/7/11Visual Aid Speech Self EvaluationVerbal Delivery:I would say that my rate, pitch, volume were all adequate. I feel that all oft he audience members could hear me, understand my pitch, and follow mywords at an easy pace without having
Lake County - SPEECH - 121
Rahul Saha7/7/11Visual Aid Speech Self EvaluationVerbal Delivery:I would say that my rate, pitch, volume were all adequate. I feel that all oft he audience members could hear me, understand my pitch, and follow mywords at an easy pace without having
CSU Long Beach - BIO - 202
Ciantera RoseHomework Set #31. Experimental Design:I would use six seedlings, 4 replicates of the plant hormone treatment and 2controls (wildtype). I would use cups and fill them with a soil media and aHoagland nutrient solution to promote growth.Ri
CSU Long Beach - BIO - 202
1.The o:p ratio in the mononitration of bromobenzene has been reported to be38:62. Using this ratio and the amount of 4-nitrobromobenzene you obtained,calculate the experimental yield of your reaction?2.Why is the bromobenzene added in portions over
USC - KOREAN - 101
G:7 . = &.IM: @ @E do v ^?G:7IM: ;= &. ; = ?G:7?=7 W@t.IM: : ~G: .!N: = !G:P & ? >?N: 7 . ; = &.IM: :N: :.:? ; = ?IM: P & ?>. ; = &.N: ; = &..IM: = h &. ; = &?N: P & ?>.G: 77W@?N: @ tEd ov.G: ; .= D ?N
Walden University - EDUC - 8110
THESIS AND TERM PAPER WRITING GUIDEFOR PSYCHOLOGY STUDENTSFourth EditionPrepared by:Dr. Robert P. Markley, Dr. Thomas T. Jackson,Marnie Arnhold & Connie Werthadapted from an original document prepared byDr. Ruby Fournet, North Texas State Universit
University of Florida - CHM - 2211L
CHEM 2211L Organic Chemistry Laboratory (2 credits)Room 238 Leigh HallSummer 2011Teaching AssistantFaculty Coordinatorto be assigned during first laboratory meetingDr. Tammy A. Davidson, Sisler 429B(352) 392-9134, davidson@chem.ufl.eduPlease use e
University of Florida - CHM - 2211L
Welcome to Organic Lab!Weve put together this module togive you important informationabout course policies and safety.Please view the presentation andread the information in your labmanual and syllabus about labsafety BEFORE you attempt thesafety
University of Florida - CHM - 2211L
Think of a problem as a puzzle, and approachit with logicCalculate the degrees of unsaturation firstDeg. Unsat = [2*C H hal + N +2]/2Look at IR for functional groupsLook at proton NMR for information aboutNMRneighboring groupsPut pieces together t
University of Florida - CHM - 2211L
NMR spectroscopy is the most powerful toolavailable for organic structure determination.It is used to study a wide variety of nuclei withspin quantum numbers (l) of , including 1H,13C, 15N, 19F, and 31P.Our focus in this lab will be primarily on 1H,
University of Florida - CHM - 2211L
When a molecule is exposed to infrared light, thebonds in the molecule begin to vibrate.The infrared region of the electromagneticspectrum is just below red in the visible region.Wavelengths are usually 2.5-25 m.More common units are wavenumbers, or
University of Florida - CHM - 2211L
Dr. Tammy DavidsonUniversity of Floridaof FloridaYou wake up in a hospital in a place you dontrecognize. You are OK no injuries or signs oftrauma, but you cant remember your name orhowhow you got there, and your wallet is gone.andHow can authorit
National-Louis - ACCT - 518
M ACROENOMICS-Week1OnJuly17,2003,TheNationalBureauofEconomicResearch(NBER)announcedthattherecessionthatbeganwithabusinesscyclepeakinMarch2001endedwithabusinesscycletroughinNovember2001.Thisannouncementhasgeneratedsomedegreeofcriticismbaseduponthefactt
University of Phoenix - BUS - 219
Running head: AMAZON.COM EVOLUTIONAmazon.com EvolutionTeam DBIS/219June 30, 2011Steven Waibel1AMAZON.COM EVOLUTION2Amazon.com EvolutionA study conducted in 2008 showed that 66% of Americans have purchased a product online(PewResearchCenter Publ
UMass (Amherst) - FIN - 473
FIN 470 Exam 11.NAME_What are the three types of financial management decisions? For each type of decision, give an exampleof a business transaction that would be relevant.Capital budgeting (deciding whether to expand a manufacturing plant), capital