Appendix C Pipeline implementation Pipeline hazards, detection and forwarding Multiple-cycle operations MIPS R4000 CDA5155 Fall 2014, Peir / University of Florida
Limits of Pipelining Increasing the number of pipeline stages in a given logic block by a factor of n generally allows increasing clock speed & throughput by a factor of almost n. Usually less than n because of overheads such as latches and balance of delay in each stage. But, pipelining has a natural limit: At least 1 layer of logic gates per pipeline stage! Practical minimum is usally several gates (2-10). Commercial designs are approaching this point!!
Simple RISC Datapath
Basic RISC Pipelining Basic idea: Each instruction spends 1 clock cycle in each of the 5 execution stages. During 1 clock cycle, the pipeline can be processing (different stages of) 5 different instructions.
Adding Pipeline Registers
Pipeline Hazards Hazards are circumstances which may lead to stalls (delays, “bubbles”) in the pipeline if not addressed. Three major types: Structural hazards: Lack of HW resources to keep all instructions moving. Data hazards Data results of earlier instrs. not yet avail. when needed. Control hazards Control decisions resulting from earlier instrs. (branches) not yet made; don’t know which new instrs. to execute.
Structural Hazard Example Suppose you had a combined instruction+data memory with only 1 read port
Hazards Produce “Bubbles”
Another View
Example Data Hazard
Forwarding for Data Hazards
Another Forwarding Example
Three Types of Data Hazards Let i be an earlier instruction, j a later one. RAW (read after write) j tries to read a value before i writes it WAW (write after write) i and j write to same place, but in the wrong order. Only occurs if >1 pipeline stage can write. WAR (write after read) j writes a new value to a location before i has read the old one. Only occurs if writes can happen before reads in pipeline.
An Unavoidable Stall - Load
Stalling for Load Dependent
Data Hazard Prevention A clever compiler can often reschedule instructions (code motion) to avoid a stall. A simple example: Original code: lw r2, 0(r4) add r1, r2, r3 Note: Stall happens here! lw r5, 4(r4) Transformed code: lw r2, 0(r4) lw r5, 4(r4) add r1, r2, r3 No stall needed!
MIPS Instruction Format
5-Stage Pipeline
Operations of Pipe Stages
Data Hazard Detection
Hazard Detection Logic for Load NOTE, The right part of the equ. should be IF/ID.IR (Fig. C.25) Example: Detecting whether an instruction that has just been fetched needs to be stalled because of dependence from a preceding load.
Forwarding Situations in MIPS Same as Figure C.26
Forwarding to The ALU Provide multiple path to the input of the ALU
Datapath with Forwarding Hardware PCSrc Read Address Instruction Memory Add PC 4 Write Data Read Addr 1 Read Addr 2 Write Addr Register File Data 1 Data 2 16 32 ALU Shift left 2 Data IF/ID Sign Extend ID/EX EX/MEM MEM/WB Control cntrl Branch Forward Unit For lecture. How many bits wide is each pipeline register now? ID/EX – 9 + 32x4 + 10 = 147 + 10 = 157 Control line inputs to Forward Unit EX/MEM.RegWrite and MEM/WB.RegWrite not shown on diagram EX/MEM.RegisterRd MEM/WB.RegisterRd ID/EX.RegisterRt ID/EX.RegisterRs
Adding the Hazard Hardware PCSrc ID/EX.MemRead Hazard Unit ID/EX ID/EX.RegisterRt EX/MEM IF/ID 1 Control Add MEM/WB Branch Add 4 Shift left 2 Instruction Memory Read Addr 1 Data Memory Register File Read Data 1 Read Addr 2 Read Address PC Read Data Address Write Addr ALU Read Data 2 Write Data Write Data ALU cntrl For lecture In reality, only the signals RegWrite and MemWrite need to be 0, the other control signals can be don’t cares. Another consideration is energy – where clock gating is called for. 16 32 Sign Extend Forward Unit
Branch Hazard Suppose the new PC value is not computed until the MEM stage. Then we must stall 3 clocks after every branch!
Early Branch Resolution Branch resolution at ID stage
Predict-Not-Taken (Branch resolves in ID) Same as Fig. C.12
Branch is taken (if taken) at this point Delayed Branches Machine code sequence: Branch instruction Delay slot instruction(s) Post-branch instructions Branch is taken (if taken) at this point Same as Fig. C.13
Filling the Branch-Delay Slot For (b), (c) must no side-effect! Note, dynamic branch prediction will be covered in Chap. 3
Multi-Cycle Execution Figure C.33 The MIPS pipeline with three additional unpipelined, floating-point, functional units.
Latency & Initiation Interval Extra delay cycles before result is available. Initiation interval: Minimum number of cycles before a new input can be given to that functional unit.
Pipelined Multiple-FP Operations Figure C.35 A pipeline that supports multiple outstanding FP operations.
Pipelining FP Instructions Notice instructions may complete out-of-order: MULTD IF ID M1 M2 M3 M4 M5 M6 M7 ME WB ADDD IF ID A1 A2 A3 A4 ME WB LD IF ID EX ME WB SD IF ID EX ME WB Raises the possibility of WAW hazards, and structural hazards in MEM & WB stages. Structural hazards may occur especially often with non-pipelined DIV unit. Out-of-order completion impacts exception handling.
Issues in Multi-Cycle Operations Stall for RAW is longer and more frequent (Fig. C.37) WAW is possible; WAR is not (why?) Structural Hazard possible for non-pipelined unit Multiple WBs are likely (Fig. C.38) Handling hazards At Issue (ID) stage: Check structural hazards: functional unit, WB port Check RAW hazards: Issue with forwarding Check WAW hazards: Not issue to make sure write in order Detect and stall instruction before MEM and WB stages More uniform handling given in Chapter 3.
Maintaining Precise Exception Settle for imprecise exception Buffer and complete in order Require large buffers and comparators History file, future file approaches Software trap handling when exception occurs Hybrid scheme: Issue when certain no exception for early instruction All instructions before can be completed No instructions after can be completed
Real MIPS R4000 Pipeline IF,IS - Instruction cache fetch, First & Second halves. RF - Inst. decode, Register Fetch, hazard check… EX - Execution (EA calc, ALU op, target calc…) DF,DS - Data cache access, First & Second halves. TC - Tag Check, did cache access hit? Note, use data before resolving hit/miss. WB - Write-Back for loads & register-register ops. Read through C.43 – C.51
2-Cycle Load Delay
Branch Delay