Chapter 4 (Part I) The Processor. Baback Izadi Division of Engineering Programs

Transcription

1 EGC442 Introdction to Compter Architectre Chapter 4 (Part I) The Processor Baback Izadi Division of Engineering Programs bai@engr.newpaltz.ed Introdction CPU performance factors Instrction cont Determined by ISA and compiler CPI and Cycle time Determined by CPU hardware We will eamine two IPS implementations A simplified version A more realistic pipelined version Simple sbset, shows most aspects emory reference: lw, sw Arithmetic/logical: add, sb, and, or, slt Control transfer: beq, j 4. Introdction 2 Division of Engineering Programs,

2 EGC442 Introdction to Compter Architectre Abstract / Simplified View Register # PC Address Instrction Registers Register # Instrction Register # Address Two types of fnctional nits: Elements that operate on vales (combinational) Elements that contain state (seqential) Registers and 3 EGC432 SUNY 3 New Paltz Instrction Eection PC instrction, fetch instrction Register nmbers register file, read registers Depending on instrction class Use to calclate Register # Address Registers Arithmetic reslt Register # Instrction emory address for load/store Register # Branch target address Access for load/store PC target address or PC + 4 PC Instrction Address 4 Division of Engineering Programs, 2

3 EGC442 Introdction to Compter Architectre ore Detailed CPU Overview ltipleers Can t jst join wires together Use mltipleers Division of Engineering Programs, 3

4 EGC442 Introdction to Compter Architectre Control Logic Design Basics Information encoded in binary Low voltage =, High voltage = One wire per bit lti-bit encoded on mlti-wire bses Combinational element Operate on Otpt is a fnction of inpt State (seqential) elements Store information 8 Division of Engineering Programs, 4

5 EGC442 Introdction to Compter Architectre Combinational Elements AND-gate Y = A & B Adder Y = A + B A B + Y A B I I S Y ltipleer Y = S? I : I Y Arithmetic/Logic Unit Y = F(A, B) 9 A B F Y Seqential Elements Register: stores in a circit Uses a clock signal to determine when to pdate the stored vale Edge-triggered: pdate when Clk changes from to D Clk Q Clk D Q Division of Engineering Programs, 5

6 EGC442 Introdction to Compter Architectre Seqential Elements Register with write control Only pdates on clock edge when write control inpt is Used when stored vale is reqired later Clk D Clk Q D Q Clocking ethodology Combinational logic transforms dring clock cycles Between clock edges Inpt from state elements, otpt to state element Longest delay determines clock period 2 Division of Engineering Programs, 6

7 EGC442 Introdction to Compter Architectre Bilding a path path Elements that process and addresses in the CPU Registers, s, m s, memories, We will bild a IPS path incrementally Refining the overview design 4.3 Bilding a path 3 Abstraction ake sre yo nderstand the abstractions! Sometimes it is easy to think yo do, when yo don t Select A 32 B 32 Select 32 C A3 B3 A3 B3. C3 C3. A B C 4 Division of Engineering Programs, 7

8 EGC442 Introdction to Compter Architectre Register File operation sing D flip-flops and UX s register nmber Register register nmber register nmber 2 register Register file 2 register nmber2 Register... Register n 2 Register n 2 What is the fnction of above? 5 Register File and register nmber register nmber register nmber 2 register Register Register file 2 Register n-to- decoder n n C Register D C Register D register nmber 2 Register Register n Register n C Register n D 2 C Register n D How many registers can we read and write at the same time? Does this spport IPS instrctions reqirements? 6 Division of Engineering Programs, 8

9 EGC442 Introdction to Compter Architectre Bilding the path Inclde the fnctional nits we need for each instrction Instrction address em Instrction Instrction PC Add Sm Address 6 Sign 32 etend Register nmbers a. Instrction b. Program conter 5 3 register 5 5 register 2 Registers register 2 Reg c. Adder control Zero reslt a. Registers b. em a. nit Control: AND OR add sbtract set-on-less-than Use mltipleors to stitch them together b. Sign-etension nit 7 Instrction Fetch 32-bit register Increment by 4 for net instrction Division of Engineering Programs, 9

10 EGC442 Introdction to Compter Architectre R-Format Instrctions two register operands Perform arithmetic/logical operation register reslt Register nmbers 5 3 register 5 5 register 2 Registers register 2 control Zero reslt Reg a. Registers b. 9 R-Format Instrctions two register operands Perform arithmetic/logical operation register reslt Instrction re gister re gister 2 Registers re gister 2 Reg 3 control Zero reslt 2 Division of Engineering Programs,

11 EGC442 Introdction to Compter Architectre Load/Store Instrctions lw $t, offset_vale($t2) or sw $t, offset_vale ($t2) register operands Calclate address sing 6-bit offset Use, bt sign-etend offset Load: and pdate register Store: register vale to 2 path for lw and sw Instrctions lw $t, offset_vale($t2) or Compte address Sign etend 6 bit to 32 bit sw $t, offset_vale ($t2) Instrction register register 2 register Reg Registers da ta Sign etend 3 operation em Zero Rea reslt Address da ta em 22 Division of Engineering Programs,

12 EGC442 Introdction to Compter Architectre Branch Instrctions PC + 4 Net instrction address beq $t, $t2, offset register operands Compare operands Instrction Use to compare registers, Registers register register 2 register Reg 6 32 Add reslt Sign etend Use to affect Z flag If not eqal, net PC <= PC +4 (Already calclated by instrction fetch) If eqal, Calclate target address Sign-etend displacement Shift left 2 places (word displacement) Add to PC Shift left 2 3 operation Zero Branch control Branch Instrctions Jst re-rotes wires Sign-bit wire replicated Division of Engineering Programs, 2

13 EGC442 Introdction to Compter Architectre Composing the Elements First-ct path does an instrction in one clock cycle Each path element can only do one fnction at a time Hence, we need separate instrction and memories Use mltipleers where alternate sorces are sed for different instrctions 25 R-Type/Load/Store path Division of Engineering Programs, 3

14 EGC442 Introdction to Compter Architectre Fll path Three Instrction Classes R-Type Instrction op rs rt rd shamt fnct rd: destination Load and Store Instrctions op rs rt 6 bit offset rt: destination Branch Instrction op rs rt 6 bit offset rt: destination 28 Division of Engineering Programs, 4

15 EGC442 Introdction to Compter Architectre Completed Path PCSrc 4 Add Shift left 2 Add reslt Reg PC address Instrction Instrction [3 ] Instrction [25 2] Instrction [2 6] Instrction [5 ] Instrction [5 ] RegDst register register 2 Registers 2 register 6 32 Sign etend Src Zero reslt Control Address em em emtoreg 29 The ain Control Unit Control signals derived from instrction R-type Load/ Store Branch rs rt rd shamt fnct 3:26 25:2 2:6 5: :6 5: 35 or 43 rs rt address 3:26 25:2 2:6 5: 4 rs rt address 3:26 25:2 2:6 5: opcode always read read, ecept for load write for R-type and load sign-etend and add 3 Division of Engineering Programs, 5

16 EGC442 Introdction to Compter Architectre path With Control R-Type Instrction Division of Engineering Programs, 6

17 EGC442 Introdction to Compter Architectre Load Instrction Branch-on-Eqal Instrction Division of Engineering Programs, 7

18 EGC442 Introdction to Compter Architectre Control Using the op-code from the instrction, the control isses signals to: Selecting the operations to perform (, read/write, etc.) Controlling the flow of (mltipleer inpts) 's operation based on instrction type and fnction code Eample: what shold the do with the instrction add $8, $7, $8 op rs rt rd shamt fnct 35 Control Eample: what shold the do with the instrction lw $, ($2) 35 2 op rs rt 6 bit offset Why is the code for sbtract is and not? 36 Division of Engineering Programs, 8

19 EGC442 Introdction to Compter Architectre Control st describe hardware to compte 3-bit control inpt Given instrction type = lw, sw Op = beq, compted from instrction type = arithmetic Fnction code for arithmetic Op Fnct field Operation Op Op F5 F4 F3 F2 F F X X X X X X add X X X X X X sb (X) X X add (X) X X sb (X) X X and (X) X X or (X) X X slt Control: AND OR add sbtract set-on-less-than 37 Control Inpts Otpts Op Fnct field Operation Op Op F5 F4 F3 F2 F F 2 X X X X X X add X X X X X X sb X X add X X sb X X and X X or X X slt li-level decoding can redce size of control nit and increase its speed. Op control block Op Op F3 F2 Operation2 Operation Operation F(5 ) F Operation F 38 Division of Engineering Programs, 9

20 EGC442 Introdction to Compter Architectre A Simple Control PCSrc Add reslt Add Shift left 2 4 Reg PC address Instrction [3 ] Instrction Instrction [25 2] Instrction [2 6] Instrction [5 ] RegDst register register 2 Registers 2 register Src Zero reslt em Address emtoreg Instrction [5 ] 6 32 Sign etend em control Instrction [5 ] op 39 Control Add reslt 4 Add RegDst Branch Shift left 2 em Instrction [3 26] emtoreg Control Op em Src Reg PC address Instrction [3 ] Instrction Instrction [25 2] Instrction [2 6] Instrction [5 ] register register 2 Registers register 2 Zero reslt Address Instrction [5 ] 6 32 Sign etend control Instrction [5 ] emto- Reg em em Instrction Opcode RegDst Src Reg Branch OpOp R-format lw sw X X beq X X 4 Division of Engineering Programs, 2

21 EGC442 Introdction to Compter Architectre Single Cycle Control Simple combinational logic Op control block Op Op Inpts Op5 Op4 Op3 Op2 Op Op F3 Operation2 Operation Otpts F2 Operation R-format Iw sw beq F(5 ) RegDst F Operation Src F emtoreg Reg em em Branch Op OpO 4 Or Simple Control Strctre All of the logic is combinational We wait for everything to settle down, and the right thing to be done might not prodce right answer right away we se write signals along with clock to determine when to write Cycle time determined by length of the longest path Content of PC Combinational logic Content of PC Clock 42 Division of Engineering Programs, 2

22 EGC442 Introdction to Compter Architectre Single Cycle Implementation Calclate cycle time assming negligible delays ecept: (2ps), and adders (2ps), register file access (ps) PCSrc Add 4 Reg Shift left 2 Add reslt PC address Instrction [3 ] Instrction Instrction [25 2] Instrction [2 6] Instrction [5 ] RegDst Instrction[5 ] register register 2 register 2 Registers 6 Sign 32 etend Src control Zero reslt em Address em emtoreg Instrction [5 ] Op 43 Single Cycle Implementation Calclate cycle time assming negligible delays ecept: (2ns), and adders (2ns), register file access (ns) PCSrc 4 Add Reg Shift left 2 Add res lt PC address In s tr ction [3 ] Instrction In strctio n [25 2 ] In strctio n [2 6] In strctio n [5 ] RegDst In strctio n [5 ] register register 2 register d ata d ata 2 Registers 6 Sign 32 etend Src control Zero reslt em Address em emtoreg Instr ction [5 ] Op Instrction Instr. emory Register Op. emory Reg. Total R-format 2ps ps 2ps ps 6ps lw 2ps ps 2ps 2ps ps 8ps sw 2ps ps 2ps 2ps 7ps beq 2ps ps 2ps 5ps 44 Division of Engineering Programs, 22

23 EGC442 Introdction to Compter Architectre Implementing Jmps Jmp 2 address 3:26 25: Jmp ses word address Update PC with concatenation of Top 4 bits of old PC 26-bit jmp address Need an etra control signal decoded from opcode 45 path With Jmps Added Division of Engineering Programs, 23

24 EGC442 Introdction to Compter Architectre Performance Isses Longest delay determines clock period Critical path: load instrction Instrction register file register file Not feasible to vary period for different instrctions Violates design principle aking the common case fast We will improve performance by pipelining 47 Division of Engineering Programs, 24