Pipeline Hazards. See P&H Chapter 4.7. Hakim Weatherspoon CS 3410, Spring 2013 Computer Science Cornell University

Transcription

1 Pipeline Hazards See P&H Chapter 4.7 Hakim Weatherspoon CS 341, Spring 213 Computer Science Cornell niversity

2 Goals for Today Data Hazards Revisit Pipelined Processors Data dependencies Problem, detection, and solutions (delaying, stalling, forwarding, bypass, etc) Hazard detection unit Forwarding unit Next time Control Hazards What is the next instruction to execute if a branch is taken? Not taken?

3 Simplicity favors regularity 32 bit instructions Smaller is faster Small register file IPS Design Principles ake the common case fast Include support for constants Good design demands good compromises Support for different type of interpretations/classes

4 Recall: IPS instruction formats All IPS instructions are 32 bits long, has 3 formats R-type op rs rt rd shamt func 6 bits 5 bits 5 bits 5 bits 5 bits 6 bits I-type op rs rt immediate 6 bits 5 bits 5 bits 16 bits J-type op immediate (target address) 6 bits 26 bits

5 Recall: IPS Instruction Types Arithmetic/Logical R-type: result and two source registers, shift amount I-type: 16-bit immediate with sign/zero extension emory Access load/store between registers and ory word, half-word and byte operations Control flow conditional branches: pc-relative addresses jumps: fixed offsets, register absolute

6 Recall: IPS Instruction Types Arithmetic/Logical ADD, ADD, SB, SB, AND, OR, OR, NOR, SLT, SLT ADDI, ADDI, ANDI, ORI, ORI, LI, SLL, SRL, SLLV, SRLV, SRAV, SLTI, SLTI LT, DIV, FLO, TLO, FHI, THI emory Access LW, LH, LB, LH, LB, LWL, LWR SW, SH, SB, SWL, SWR Control flow BEQ, BNE, BLEZ, BLTZ, BGEZ, BGTZ J, JR, JAL, JALR, BEQL, BNEL, BLEZL, BGTZL Special LL, SC, SYSCALL, BREAK, SYNC, COPROC

7 Pipelined Processor ory register file alu +4 addr PC new pc control extend compute jump/branch targets d in d out ory Fetch Decode Execute emory WB

8 ctrl ctrl ctrl inst imm B A B D D Pipelined Processor ory register file alu +4 addr PC new pc control extend compute jump/branch targets d in d out ory Instruction Fetch Instruction Decode Execute emory Write- Back IF/ID ID/E E/E E/WB

9 Example: : Sample Code (Simple) add r3, r1, r2; nand r6, r4, r5; lw r4, 2(r2); add r5, r2, r5; sw r7, 12(r3);

10 Example: Sample Code (Simple) Assume eight-register machine Run the following code on a pipelined datapath add r3 r1 r2 ; reg 3 = reg 1 + reg 2 nand r6 r4 r5 ; reg 6 = ~(reg 4 & reg 5) lw r4 2 (r2) ; reg 4 = em[reg2+2] add r5 r2 r5 ; reg 5 = reg 2 + reg 5 sw r7 12(r3) ; em[reg3+12] = reg 7 Slides thanks to Sally ckee

11 Clock cycle Time Graphs add nand lw add sw IF ID E E WB IF ID E E WB IF ID E E WB IF ID E E WB IF ID E E WB Latency: Throughput: Concurrency: CPI =

12 Register file PC 4 + Inst PC+4 instruction rega regb Bits Bits 16-2 Bits R R1 R2 R3 R4 R5 R6 R7 extend PC+4 vala valb imm Rd Rt op A L target AL result valb dest op Data AL result mdata IF/ID ID/E E/E E/WB dest op data dest

13 At time 1, Fetch add r3 r1 r2 extend data dest IF/ID ID/E E/E E/WB

14 Register file add PC 4 Fetch: add Inst 4 add Bits Bits 16-2 Bits R R1 R2 R3 R4 R5 R6 R extend / 4 / 36 / 9 / 3 / 2 nop / add IF/ID ID/E E/E E/WB Time: 1 / 2 A L nop Data nop data dest

15 Register file nand add PC 4 Fetch: nand Inst 8 nand Bits Bits 16-2 Bits R R1 R2 R3 R4 R5 R6 R extend / / 18 / add / nand A L / 4 / 45 / 9 / 36 3 / 25 / 3 nop / add IF/ID ID/E E/E E/WB Time: 2 / 3 Data nop data dest

16 Register file lw 4 2(2) nand add nand (18 7) PC 4 Fetch: lw 4 2(2) Time: 3 / 4 + Inst 12 lw 4 2(2) 4 5 Bits Bits 16-2 Bits R R1 R2 R3 R4 R5 R6 R extend nand = = = / 18 / 9 7 A L 3 / / -3 / 9 7 / 3 6 add / nand Data / 45 / 3 nop / add IF/ID ID/E E/E E/WB data dest

17 Register file add lw 4 2(2) nand add PC 4 Fetch: add Time: 4 + Inst 16 add Bits Bits 16-2 Bits R R1 R2 R3 R4 R5 R6 R extend lw A L nand Data IF/ID ID/E E/E E/WB add data dest

18 Register file sw 7 12(3) add lw 4 2 (2) nand add PC 4 Fetch: sw 7 12(3) Time: 5 + Inst 2 sw 7 12(3) 2 5 Bits Bits 16-2 Bits R R1 R2 R3 R4 R5 R6 R extend add A L lw -3 6 Data -3 6 nand IF/ID ID/E E/E E/WB 45 3 data dest

19 Register file sw 7 12(3) add lw 4 2(2) nand PC 4 No more instructions Time: 6 + Inst 3 7 Bits Bits 16-2 Bits R R1 R2 R3 R4 R5 R6 R extend sw A L add Data IF/ID ID/E E/E E/WB lw -3 6 data dest

20 Register file nop nop sw 7 12(3) add lw 4 2(2) PC 4 No more instructions Time: 7 + Inst Bits Bits 16-2 Bits R R1 R2 R3 R4 R5 R6 R extend A L sw Data IF/ID ID/E E/E E/WB add 99 4 data dest

21 Register file nop nop nop sw 7 12(3) add PC 4 + Inst R R1 R2 R3 R4 R5 R6 R extend A L Data data dest No more instructions Time: 8 Bits Bits 16-2 Bits IF/ID ID/E E/E E/WB 7 sw 5 Slides thanks to Sally ckee

22 Register file nop nop nop nop sw 7 12(3) PC 4 + Inst R R1 R2 R3 R4 R5 R6 R extend A L Data data dest No more instructions Bits Bits 16-2 Bits Time: 9 IF/ID ID/E E/E E/WB

23 Takeaway Pipelining is a powerful technique to mask latencies and increase throughput Logically, instructions execute one at a time Physically, instructions execute in parallel Instruction level parallelism Abstraction promotes decoupling Interface (ISA) vs. implementation (Pipeline)

24 Next Goal What about data dependencies (also known as a data hazard in a pipelined processor)? i.e. add r3, r1, r2 sub r5, r3, r4

25 Data Hazards Data Hazards register file reads occur in stage 2 (ID) register file writes occur in stage 5 (WB) next instructions may read values about to be written

26 time Data Hazards Clock cycle add r3, r1, r2 IF ID E WB sub r5, r3, r4 IF ID E WB lw r6, 4(r3) IF ID E WB or r5, r3, r5 IF ID E WB sw r6, 12(r3) IF ID E WB

27 Data Hazards Data Hazards register file reads occur in stage 2 (ID) register file writes occur in stage 5 (WB) next instructions may read values about to be written How to detect?

28 inst PC+4 Detecting Data Hazards OP Rt Rd PC+4 imm B A B D OP Rd D OP Rd add r3, r1, r2 sub inst r5, r3, r5 or r6, r3, r4 add r6, r3, r8 A Rd D B Ra Rb addr PC +4 detect hazard d in d out IF/ID ID/E E/E E/WB

29 Takeaway Data hazards occur when a operand (register) depends on the result of a previous instruction that may not be computed yet. A pipelined processor needs to detect data hazards.

30 Next Goal What to do if data hazard detected?

31 Stalling How to stall an instruction in ID stage prevent IF/ID pipeline register update stalls the ID stage instruction convert ID stage instr into nop for later stages innocuous bubble passes through pipeline prevent PC update stalls the next (IF stage) instruction

32 inst PC+4 Detecting Data Hazards OP Rt Rd PC+4 imm B A B D OP Rd D OP Rd add r3, r1, r2 sub inst r5, r3, r5 or r6, r3, r4 add r6, r3, r8 A Rd D B Ra Rb addr PC +4 detect hazard d in d out If detect hazard WE= IF/ID emwr= RegWr= ID/E E/E E/WB

33 time Stalling Clock cycle add r3, r1, r2 sub r5, r3, r5 or r6, r3, r4 add r6, r3, r8

34 time r3 = 1 add r3, r1, r2 r3 = 2 Stalling Clock cycle sub r5, r3, r5 or r6, r3, r4 add r6, r3, r8

35 inst Stalling Rd WE Op Rd WE Op Rd WE Op inst +4 D rd ra rb A B A B D B data D PC (emwr= RegWr=) nop sub r5,r3,r5 add r3,r1,r2 or r6,r3,r4 (WE=) /stall NOP = If(IF/ID.rA && (IF/ID.rA==ID/Ex.Rd IF/ID.rA==Ex/.Rd IF/ID.rA==/W.Rd))

36 inst Stalling Rd WE Op Rd WE Op Rd WE Op inst +4 D rd ra rb A B A B D B data D PC (emwr= RegWr=) nop sub r5,r3,r5 (emwr= RegWr=) nop add r3,r1,r2 or r6,r3,r4 (WE=) /stall NOP = If(IF/ID.rA && (IF/ID.rA==ID/Ex.Rd IF/ID.rA==Ex/.Rd IF/ID.rA==/W.Rd))

37 inst Stalling Rd WE Op Rd WE Op Rd WE Op inst +4 D rd ra rb A B A B D B data D PC or r6,r3,r4 (WE=) (emwr= RegWr=) nop sub r5,r3,r5 nop nop /stall NOP = If(IF/ID.rA && (IF/ID.rA==ID/Ex.Rd IF/ID.rA==Ex/.Rd IF/ID.rA==/W.Rd)) (emwr= RegWr=) (emwr= RegWr=) add r3,r1,r2

38 Stalling How to stall an instruction in ID stage prevent IF/ID pipeline register update stalls the ID stage instruction convert ID stage instr into nop for later stages innocuous bubble passes through pipeline prevent PC update stalls the next (IF stage) instruction

39 Takeaway Data hazards occur when a operand (register) depends on the result of a previous instruction that may not be computed yet. A pipelined processor needs to detect data hazards. Stalling, preventing a dependent instruction from advancing, is one way to resolve data hazards. Stalling introduces NOPs ( bubbles ) into a pipeline. Introduce NOPs by (1) preventing the PC from updating, (2) preventing writes to IF/ID registers from changing, and (3) preventing writes to ory and register file. Bubbles in pipeline significantly decrease performance.

40 ext Goal: Resolving Data Hazards via Forwarding What to do if data hazard detected? A) Wait/Stall B) Reorder in Software (SW) C) Forward/Bypass

41 Forwarding Forwarding bypasses some pipelined stages forwarding a result to a dependent instruction operand (register). Three types of forwarding/bypass Forwarding from Ex/em registers to Ex stage ( Ex) Forwarding from em/wb register to Ex stage (W Ex) RegisterFile Bypass

42 Forwarding Datapath imm Rb Ra Rd WE C Rd WE C inst D A B A B D B data D detect hazard forward unit IF/ID ID/Ex Ex/em em/wb Three types of forwarding/bypass Forwarding from Ex/em registers to Ex stage ( Ex) Forwarding from em/wb register to Ex stage (W Ex) RegisterFile Bypass

43 Forwarding Datapath imm Rb Ra Rd WE C Rd WE C inst D A B A B D B data D detect hazard forward unit IF/ID ID/Ex Ex/em em/wb Three types of forwarding/bypass Forwarding from Ex/em registers to Ex stage ( Ex) Forwarding from em/wb register to Ex stage (W Ex) RegisterFile Bypass

44 Forwarding Datapath 1 Ex/E to E Bypass E needs AL result that is still in E stage Resolve: Add a bypass from E/E.D to start of E How to detect? Logic in Ex Stage: forward = (Ex/.WE && E/.Rd!= && ID/Ex.Ra == Ex/.Rd) (same for Rb)

45 Forwarding Datapath 1 inst D A B data add r3, r1, r2 sub r5, r3, r1 IF ID Ex W IF ID Ex W

46 Forwarding Datapath 2 em/wb to E Bypass E needs value being written by WB Resolve: Add bypass from WB final value to start of E How to detect? Logic in Ex Stage: forward = (/WB.WE && /WB.Rd!= && ID/Ex.Ra == /WB.Rd && not (ID/Ex.WE && Ex/.Rd!= && ID/Ex.Ra == Ex/.Rd) (same for Rb) Check pg. 369

47 Forwarding Datapath 2 inst D A B data add r3, r1, r2 IF ID Ex W sub r5, r3, r1 or r6, r3, r4 IF ID IF Ex W ID Ex W

48 Register File Bypass Register File Bypass Reading a value that is currently being written Detect: Resolve: ((Ra == E/WB.Rd) or (Rb == E/WB.Rd)) and (WB is writing a register) Add a bypass around register file (WB to ID) Better: (Hack) just negate register file clock writes happen at end of first half of each clock cycle reads happen during second half of each clock cycle

49 Register File Bypass inst D A B data add r3, r1, r2 IF ID Ex W sub r5, r3, r1 or r6, r3, r4 add r6, r3, r8 IF ID IF Ex W ID Ex W IF ID Ex W

50 time r3 = 1 add r3, r1, r2 r3 = 2 Forwarding Example Clock cycle sub r5, r3, r5 or r6, r3, r4 add r6, r3, r8

51 time add r3, r1, r2 Forwarding Example 2 Clock cycle IF ID Ex W sub r5, r3, r4 IF ID Ex W lw r6, 4(r3) or r5, r3, r5 sw r6, 12(r3)

52 Tricky Example inst D A B data add r1, r1, r2 SB r1, r1, r3 OR r1, r4, r1

53 Forwarding Datapath imm Rb Ra Rd WE C Rd WE C inst D A B A B D B data D detect hazard forward unit IF/ID ID/Ex Ex/em em/wb Three types of forwarding/bypass Forwarding from Ex/em registers to Ex stage ( Ex) Forwarding from em/wb register to Ex stage (W Ex) Register File Bypass

54 Takeaway Data hazards occur when a operand (register) depends on the result of a previous instruction that may not be computed yet. A pipelined processor needs to detect data hazards. Stalling, preventing a dependent instruction from advancing, is one way to resolve data hazards. Stalling introduces NOPs ( bubbles ) into a pipeline. Introduce NOPs by (1) preventing the PC from updating, (2) preventing writes to IF/ID registers from changing, and (3) preventing writes to ory and register file. Bubbles (nops) in pipeline significantly decrease performance. Forwarding bypasses some pipelined stages forwarding a result to a dependent instruction operand (register). Better performance than stalling.

55 Administrivia Prelim1: next Tuesday, February 26 th in evening Time: We will start at 7:3pm sharp, so come early Prelim Review: This Thur 6-8pm in pson B14 and Fri, 5-7pm in Phillips 23 Closed Book Cannot use electronic device or outside material Practice prelims are online in CS aterial covered everything up to end of this week Appendix C (logic, gates, FSs, ory, ALs) Chapter 4 (pipelined [and non-pipeline] IPS processor with hazards) Chapters 2 (Numbers / Arithmetic, simple IPS instructions) Chapter 1 (Performance) HW1, HW2, Lab, Lab1, Lab2

56 Administrivia HW2 is due tomorrow Fill out Survey online. Receive credit/points on homework for survey: Should have received from Kathryn Dimiduk Survey is anonymous Project1 (PA1) due week after prelim Continue working diligently. se design doc momentum Save your work! Save often. Verify file is non-zero. Periodically save to Dropbox, . Beware of acos 1.5 (leopard) and 1.6 (snow-leopard) se your resources Lab Section, Piazza.com, Office Hours, Homework Help Session, Class notes, book, Sections, CSGLab

57 Administrivia Check online syllabus/schedule Slides and Reading for lectures Office Hours Homework and Programming Assignments Prelims (in evenings): Tuesday, February 26 th Thursday, arch 28 th Thursday, April 25 th Schedule is subject to change

58 Collaboration, Late, Re-grading Policies Black Board Collaboration Policy Can discuss approach together on a black board Leave and write up solution independently Do not copy solutions Late Policy Each person has a total of four slip days ax of two slip days for any individual assignment Slip days deducted first for any late assignment, cannot selectively apply slip days For projects, slip days are deducted from all partners 25% deducted per day late after slip days are exhausted Regrade policy Submit written request to lead TA, and lead TA will pick a different grader Submit another written request, lead TA will regrade directly Submit yet another written request for professor to regrade.

59 Quiz Find all hazards, and say how they are resolved: add r3, r1, r2 sub r3, r2, r1 nand r4, r3, r1 or r, r3, r4 xor r1, r4, r3 sb r4, 1(r)

60 emory Load Data Hazard inst D A B data lw r4, 2(r8) sub r6, r4, r1

61 Resolving emory Load Hazard Load Data Hazard Value not available until WB stage So: next instruction can t proceed if hazard detected Resolution: IPS 2/3: one delay slot ISA says results of loads are not available until one cycle later Assembler inserts nop, or reorders to fill delay slot IPS 4 onwards: stall But really, programmer/compiler reorders to avoid stalling in the load delay slot

62 Quiz 2 add r3, r1, r2 nand r5, r3, r4 add r2, r6, r3 lw r6, 24(r3) sw r6, 12(r2)

63 Data Hazard Recap Delay Slot(s) odify ISA to match implementation Stall Pause current and all subsequent instructions Forward/Bypass Try to steal correct value from elsewhere in pipeline Otherwise, fall back to stalling or require a delay slot Tradeoffs?