CSEE 3827: Fundamentals of Computer Systems
Lecture 18, 19, & 20
April 2009
Martha Kim [email protected] Outline
• We will examine two MIPS implementations
• A single-cycle version
• A pipelined version
• Simple subset of MIPS, showing most aspects
• Memory reference: lw, sw
• Arithmetic/logical: add, sub, and, or, slt
• Control transfer: beq, j
• CPU performance factors
• Instruction count (determined by ISA and compiler)
• Cycles per instruction and cycle time (determined by CPU hardware)
CSEE 3827, Spring 2009 Martha Kim 2 Instruction Execution
• PC → instruction memory, fetch instruction
• Register numbers → register file, read registers
• Depending on instruction class:
• Use ALU to calculate:
• Arithmetic or logical result
• Memory address for load/store
• Branch target address
• Access data for load/store
• PC ← target address or PC + 4
CSEE 3827, Spring 2009 Martha Kim 3 CPU Overview
!DD !DD
$ATA
2EGISTER 0# !DDRESS )NSTRUCTION 2EGISTERS !,5 !DDRESS 2EGISTER $ATA )NSTRUCTION MEMORY MEMORY 2EGISTER
$ATA
1, Ê {°£Ê Ê >LÃÌÀ>VÌÊ ÛiÜÊ vÊ Ì iÊ «iiÌ>ÌÊ vÊ Ì iÊ *-Ê ÃÕLÃiÌÊ Ã Ü}Ê Ì iÊ >ÀÊvÕVÌ>ÊÕÌÃÊ>`ÊÌ iÊ>ÀÊViVÌÃÊLiÌÜiiÊÌ i°Ê!LL INSTRUCTIONS START BY USING THE PRO GRAM COUNTER TO SUPPLY THE INSTRUCTION ADDRESS TO THE INSTRUCTION MEMORY !FTER THE INSTRUCTION IS FETCHED THE REGISTER OPERANDS USED BY AN INSTRUCTION ARE SPECIl ED BY l ELDS OF THAT INSTRUCTION /NCE THE REGISTER OPERANDS HAVE BEEN FETCHED THEY CAN BE OPERATED ON TO COMPUTE A MEMORY ADDRESS FOR A LOAD OR STORE TO COMPUTE AN ARITHMETIC RESULT FOR AN INTEGER ARITHMETIC LOGICAL INSTRUCTION OR A COMPARE FOR A BRANCH )F THE INSTRUCTION IS AN ARITHMETIC LOGICAL INSTRUCTION THE RESULT FROM THE !,5 MUST BE WRITTEN TO A REGISTER )F THE OPERATION IS A LOAD OR STORE THE !,5 RESULT IS USED AS AN ADDRESS TO EITHER STORE A VALUE FROM THE REGISTERS OR LOAD A VALUE FROM MEMORY INTO THE REGISTERS 4HE RESULT FROM THE !,5 OR MEMORY IS WRITTEN BACK INTO THE REGISTER l LE "RANCHES REQUIRE THE USE OF THE !,5 OUTPUT TO DETERMINE THE NEXT INSTRUCTION ADDRESS WHICH COMES EITHER FROM THE !,5 WHERE THE 0# AND BRANCH OFFSET ARE SUMMED OR FROM AN ADDER THAT INCREMENTS THE CURRENT 0# BY 4HE THICK LINES INTERCONNECTING THE FUNCTIONAL UNITS REPRESENT BUSES WHICH CONSIST OF MULTIPLE SIGNALS 4HE ARROWS ARE USED TO GUIDE THE READER IN KNOWING HOW INFORMATION m OWS 3INCE SIGNAL LINES MAY CROSS WE EXPLICITLY SHOW WHEN CROSSING LINES ARE CONNECTED BY THE PRESENCE OF A DOT WHERE THE LINES CROSS #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 4 Can’t just join wires together, use muxes
!DD !DD
$ATA
2EGISTER 0# !DDRESS )NSTRUCTION 2EGISTERS !,5 !DDRESS 2EGISTER $ATA )NSTRUCTION MEMORY MEMORY 2EGISTER
$ATA
1, Ê {°£Ê Ê >LÃÌÀ>VÌÊ ÛiÜÊ vÊ Ì iÊ «iiÌ>ÌÊ vÊ Ì iÊ *-Ê ÃÕLÃiÌÊ Ã Ü}Ê Ì iÊ >ÀÊvÕVÌ>ÊÕÌÃÊ>`ÊÌ iÊ>ÀÊViVÌÃÊLiÌÜiiÊÌ i°Ê!LL INSTRUCTIONS START BY USING THE PRO GRAM COUNTER TO SUPPLY THE INSTRUCTION ADDRESS TO THE INSTRUCTION MEMORY !FTER THE INSTRUCTION IS FETCHED THE REGISTER OPERANDS USED BY AN INSTRUCTION ARE SPECIl ED BY l ELDS OF THAT INSTRUCTION /NCE THE REGISTER OPERANDS HAVE BEEN FETCHED THEY CAN BE OPERATED ON TO COMPUTE A MEMORY ADDRESS FOR A LOAD OR STORE TO COMPUTE AN ARITHMETIC RESULT FOR AN INTEGER ARITHMETIC LOGICAL INSTRUCTION OR A COMPARE FOR A BRANCH )F THE INSTRUCTION IS AN ARITHMETIC LOGICAL INSTRUCTION THE RESULT FROM THE !,5 MUST BE WRITTEN TO A REGISTER )F THE OPERATION IS A LOAD OR STORE THE !,5 RESULT IS USED AS AN ADDRESS TO EITHER STORE A VALUE FROM THE REGISTERS OR LOAD A VALUE FROM MEMORY INTO THE REGISTERS 4HE RESULT FROM THE !,5 OR MEMORY IS WRITTEN BACK INTO THE REGISTER l LE "RANCHES REQUIRE THE USE OF THE !,5 OUTPUT TO DETERMINE THE NEXT INSTRUCTION ADDRESS WHICH COMES EITHER FROM THE !,5 WHERE THE 0# AND BRANCH OFFSET ARE SUMMED OR FROM AN ADDER THAT INCREMENTS THE CURRENT 0# BY 4HE THICK LINES INTERCONNECTING THE FUNCTIONAL UNITS REPRESENT BUSES WHICH CONSIST OF MULTIPLE SIGNALS 4HE ARROWS ARE USED TO GUIDE THE READER IN KNOWING HOW INFORMATION m OWS 3INCE SIGNAL LINES MAY CROSS WE EXPLICITLY SHOW WHEN CROSSING LINES ARE CONNECTED BY THE PRESENCE OF A DOT WHERE THE LINES CROSS #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 5 Control
"RANCH
- U X
- !DD !DD U X
!,5 OPERATION $ATA
-EM7RITE 2EGISTER 0# !DDRESS )NSTRUCTION 2EGISTERS !,5 !DDRESS 2EGISTER - :ERO $ATA )NSTRUCTION U X MEMORY MEMORY 2EGISTER 2EG7RITE $ATA -EM2EAD
#ONTROL
1, Ê{°ÓÊ / iÊL>ÃVÊ«iiÌ>ÌÊvÊÌ iÊ *-ÊÃÕLÃiÌ]ÊVÕ`}ÊÌ iÊiViÃÃ>ÀÞÊÕÌ«iÝÀÃÊ>`ÊVÌÀÊiÃ°Ê 4HE TOP MULTIPLEXOR h-UXv CONTROLS WHAT VALUE REPLACES THE 0# 0# OR THE BRANCH DESTINATION ADDRESS THE MULTI PLEXOR IS CONTROLLED BY THE GATE THAT h!.$Sv TOGETHER THE :ERO OUTPUT OF THE !,5 AND A CONTROL SIGNAL THAT INDICATES THAT THE INSTRUC TION IS A BRANCH 4HE MIDDLE MULTIPLEXOR WHOSE OUTPUT RETURNS TO THE REGISTER l LE IS USED TO STEER THE OUTPUT OF THE !,5 IN THE CASE OF AN ARITHMETIC LOGICAL INSTRUCTION OR THE OUTPUT OF THE DATA MEMORY IN THE CASE OF A LOAD FOR WRITING INTO THE REGISTER l LE &INALLY THE BOTTOMMOST MULTIPLEXOR IS USED TO DETERMINE WHETHER THE SECOND !,5 INPUT IS FROM THE REGISTERS FOR AN ARITHMETIC LOGICAL INSTRUCTION /2 A BRANCH OR FROM THE OFFSET l ELD OF THE INSTRUCTION FOR A LOAD OR STORE 4HE ADDED CONTROL LINES ARE STRAIGHTFORWARD AND DETERMINE THE OPERATION PERFORMED AT THE !,5 WHETHER THE DATA MEMORY SHOULD READ OR WRITE AND WHETHER THE REGISTERS SHOULD PERFORM A WRITE OPERATION 4HE CONTROL LINES ARE SHOWN IN COLOR TO MAKE THEM EASIER TO SEE #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 6 MIPS Datapath Combinational Elements
• AND gate (Y = A & B) • Adder (Y = A + B)
A A Y B + Y B
• Multiplexer (Y = S ? A : B) • Arithmetic/Logic Unit (ALU) (Y = F(A,B)) A Y B A ALU Y S B
F
CSEE 3827, Spring 2009 Martha Kim 8 Clocking Methodology
Combinational logic transforms data during clock cycles. Longest combinational delay determines clock period.
3TATE 3TATE ELEMENT #OMBINATIONAL LOGIC ELEMENT
#LOCK CYCLE
1, Ê{°ÎÊ L>Ì>Ê}V]ÊÃÌ>ÌiÊiiiÌÃ]Ê>`ÊÌ iÊVVÊ>ÀiÊVÃiÞÊÀi>Ìi`°Ê)N A SYNCHRONOUS DIGITAL SYSTEM THE CLOCK DETERMINES WHEN ELEMENTS WITH STATE WILL WRITE VALUES INTO INTERNAL STORAGE !NY INPUTS TO A STATE ELEMENT MUST REACH A STABLE VALUE THAT IS HAVE REACHED A VALUE FROM WHICH THEY WILL NOT CHANGE UNTIL AFTER THE CLOCK EDGE BEFORE THE ACTIVE CLOCK EDGE CAUSES THE STATE TO BE UPDATED !LL STATE ELEMENTS IN THIS CHAPTER INCLUDING MEMORY ARE ASSUMED TO BE EDGE TRIGGERED #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
3TATE #OMBINATIONAL LOGIC ELEMENT
1, Ê{°{Ê Êi`}iÌÀ}}iÀi`ÊiÌ `}ÞÊ>ÜÃÊ>ÊÃÌ>ÌiÊiiiÌÊÌÊLiÊÀi>`Ê>`ÊÜÀÌ ÌiÊÊÌ iÊÃ>iÊVVÊVÞViÊÜÌ ÕÌÊVÀi>Ì}Ê>ÊÀ>ViÊÌ >ÌÊVÕ`Êi>`ÊÌÊ`iÌiÀ>ÌiÊ`>Ì>Ê Û>Õið /F COURSE THE CLOCK CYCLE STILL MUST BE LONG ENOUGH SO THAT THE INPUT VALUES ARE STABLE WHEN THE ACTIVE CLOCK EDGE OCCURS &EEDBACK CANNOT OCCUR WITHIN ONE CLOCK CYCLE BECAUSE OF THE EDGE TRIGGERED UPDATE OF THE STATE ELEMENT )F FEEDBACK WERE POSSIBLE THIS DESIGN COULD NOT WORK PROPERLY /UR DESIGNS IN THIS CHAPTER AND THE NEXT RELY ON THE EDGE TRIGGERED TIMING METHODOLOGY AND ON STRUCTURES LIKE THE ONE SHOWN IN THIS l GURE #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 9 Building a datapath incrementally
• Datapath: elements that process data and addresses in the CPU
• Datapath will execute one instruction in one clock cycle
• Each datapath element can only do one function at a time
• Hence, we need separate instruction and data memories
• Use multiplexers where alternate data sources are used for different instructions
CSEE 3827, Spring 2009 Martha Kim 10 Instruction Fetch
• Fetch Instruction contained in PC register from memory
• Compute PC + 4 for next instruction
)NSTRUCTION ADDRESS
)NSTRUCTION 0# !DD 3UM
)NSTRUCTION MEMORY
A )NSTRUCTION MEMORY B 0ROGRAM COUNTER C !DDER
1, Ê {°xÊ /ÜÊ ÃÌ>ÌiÊ iiiÌÃÊ >ÀiÊ ii`i`Ê ÌÊ ÃÌÀiÊ >`Ê >VViÃÃÊ ÃÌÀÕVÌÃ]Ê >`Ê >Ê >``iÀÊÃÊii`i`ÊÌÊV«ÕÌiÊÌ iÊiÝÌÊÃÌÀÕVÌÊ>``ÀiÃðÊ4HE STATE ELEMENTS ARE THE INSTRUCTION MEMORY AND THE PROGRAM COUNTER 4HE INSTRUCTION MEMORY NEED ONLY PROVIDE READ ACCESS BECAUSE THE DATAPATH DOES NOT WRITE INSTRUCTIONS 3INCE THE INSTRUCTION MEMORY ONLY READS WE TREAT IT AS COMBINATIONAL LOGIC THE OUTPUT AT ANY TIME REm ECTS THE CONTENTS OF THE LOCATION SPECIl ED BY THE ADDRESS INPUT AND NO READ CONTROL SIGNAL IS NEEDED 7E WILL NEED TO WRITE THE INSTRUCTION MEMORY WHEN WE LOAD THE PROGRAM THIS IS NOT HARD TO ADD AND WE IGNORE IT FOR SIMPLICITY 4HE PROGRAM COUNTER IS A BIT REGISTER THAT IS WRITTEN AT THE END OF EVERY CLOCK CYCLE AND THUS DOES NOT NEED A WRITE CONTROL SIGNAL 4HE ADDER IS AN !,5 WIRED TO ALWAYS ADD ITS TWO BIT INPUTS AND PLACE THE SUM ON ITS OUTPUT #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 11 Part 1: Instruction Fetch
!DD
2EAD 0# ADDRESS
)NSTRUCTION
)NSTRUCTION MEMORY
1, Ê{°ÈÊ Ê«ÀÌÊvÊÌ iÊ`>Ì>«>Ì ÊÕÃi`ÊvÀÊviÌV }ÊÃÌÀÕVÌÃÊ>`ÊVÀiiÌ}Ê Ì iÊ«À}À>ÊVÕÌiÀ°Ê4HE FETCHED INSTRUCTION IS USED BY OTHER PARTS OF THE DATAPATH #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 12 R-Format Instructions
• Read two register operands • Perform arithmetic/logical operation • Write register result
2EAD !,5 OPERATION REGISTER 2EAD 2EGISTER 2EAD DATA NUMBERS REGISTER :ERO $ATA !,5 2EGISTERS !,5 7RITE RESULT REGISTER 2EAD 7RITE DATA $ATA $ATA 2EG7RITE
A 2EGISTERS B !,5
1, Ê {°ÇÊ / iÊ ÌÜÊ iiiÌÃÊ ii`i`Ê ÌÊ «iiÌÊ ,vÀ>ÌÊ 1Ê «iÀ>ÌÃÊ >ÀiÊ Ì iÊ Ài}ÃÌiÀÊwÊiÊ>`ÊÌ iÊ 1°Ê4HE REGISTER l LE CONTAINS ALL THE REGISTERS AND HAS TWO READ PORTS AND ONE WRITE PORT 4HE DESIGN OF MULTIPORTED REGISTER l LES IS DISCUSSED IN 3ECTION # OF !PPENDIX # 4HE REGISTER l LE ALWAYS OUTPUTS THE CONTENTS OF THE REGISTERS CORRESPONDING TO THE 2EAD REGISTER INPUTS ON THE OUTPUTS NO OTHER CONTROL INPUTS ARE NEEDED )N CONTRAST A REGISTER WRITE MUST BE EXPLICITLY INDICATED BY ASSERTING THE WRITE CONTROL SIGNAL 2EMEMBER THAT WRITES ARE EDGE T RIGGERED SO THAT ALL THE WRITE INPUTS IE THE VALUE TO BE WRITTEN THE REGISTER NUMBER AND THE WRITE CONTROL SIGNAL MUST BE VALID AT THE CLOCK EDGE 3INCE WRITES TO THE REGISTER l LE ARE EDGE T RIGGERED OUR DESIGN CAN LEGALLY READ AND WRITE THE SAME REGISTER WITHIN A CLOCK CYCLE THE READ WILL GET THE VALUE WRITTEN IN AN EARLIER CLOCK CYCLE WHILE THE VALUE WRITTEN WILL BE AVAILABLE TO A READ IN A SUBSEQUENT CLOCK CYCLE 4HE INPUTS CARRYING THE REGISTER NUMBER TO THE REGISTER l LE ARE ALL BITS WIDE WHEREAS THE LINES CARRYING DATA VALUES ARE BITS WIDE 4HE OPERATION TO BE PERFORMED BY THE !,5 IS CONTROLLED WITH THE !,5 OPERATION SIGNAL WHICH WILL BE BITS WIDE USING THE !,5 DESIGNED IN !PPENDIX # 7E WILL USE THE :ERO DETECTION OUTPUT OF THE !,5 SHORTLY TO IMPLEMENT BRANCHES 4HE OVERm OW OUTPUT WILL NOT BE NEEDED UNTIL 3ECTION WHEN WE DISCUSS EXCEPTIONS WE OMIT IT UNTIL THEN #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 13 Load/Store Instructions
• Read register operands • Calculate address using 16-bit offset (use ALU but sign-extend offset) • Load: read memory and update register • Store: write register value to memory
-EM7RITE
2EAD !DDRESS DATA 3IGN $ATA EXTEND 7RITE MEMORY DATA
-EM2EAD
A $ATA MEMORY UNIT B 3IGN EXTENSION UNIT
1, Ê {°nÊ / iÊ ÌÜÊ ÕÌÃÊ ii`i`Ê ÌÊ «iiÌÊ >`ÃÊ >`Ê ÃÌÀiÃ]Ê Ê >``ÌÊ ÌÊ Ì iÊ Ài}ÃÌiÀÊwÊiÊ>`Ê 1ÊvÊ}ÕÀiÊ{°Ç]Ê>ÀiÊÌ iÊ`>Ì>ÊiÀÞÊÕÌÊ>`ÊÌ iÊÃ}ÊiÝÌiÃÊÕÌ° 4HE MEMORY UNIT IS A STATE ELEMENT WITH INPUTS FOR THE ADDRESS AND THE WRITE DATA AND A SINGLE OUTPUT FOR THE READ RESULT 4HERE ARE SEPARATE READ AND WRITE CONTROLS ALTHOUGH ONLY ONE OF THESE MAY BE ASSERTED ON ANY GIVEN CLOCK 4HE MEMORY UNIT NEEDS A READ SIGNAL SINCE UNLIKE THE REGISTER l LE READING THE VALUE OF AN INVALID ADDRESS CAN CAUSE PROBLEMS AS WE WILL SEE IN #HAPTER 4HE SIGN EXTENSION UNIT HAS A BIT INPUT THAT IS SIGN EXTENDED INTO A BIT RESULT APPEARING ON THE OUTPUT SEE #HAPTER 7E ASSUME THE DATA MEMORY IS EDGE TRIGGERED FOR WRITES 3TANDARD MEMORY CHIPS ACTUALLY HAVE A WRITE ENABLE SIGNAL THAT IS USED FOR WRITES !LTHOUGH THE WRITE ENABLE IS NOT EDGE TRIGGERED OUR EDGE TRIGGERED DESIGN COULD EASILY BE ADAPTED TO WORK WITH REAL MEMORY CHIPS 3EE 3ECTION # OF !PPENDIX # FOR FURTHER DISCUSSION OF HOW REAL MEMORY CHIPS WORK #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 14 Part 2: R-Type/Load/Store Datapath
2EAD !,5 OPERATION REGISTER 2EAD -EM7RITE DATA 2EAD :ERO -EMTO2EG )NSTRUCTION REGISTER !,53RC !,5 2EGISTERS 2EAD !,5 2EAD 7RITE !DDRESS DATA RESULT DATA REGISTER - - U U X X 7RITE DATA $ATA 7RITE MEMORY 2EG7RITE DATA
3IGN -EM2EAD EXTEND
1, Ê{°£äÊ / iÊ`>Ì>«>Ì ÊvÀÊÌ iÊiÀÞÊÃÌÀÕVÌÃÊ>`ÊÌ iÊ,ÌÞ«iÊÃÌÀÕVÌðÊ4HIS EXAMPLE SHOWS HOW A SINGLE DATAPATH CAN BE ASSEMBLED FROM THE PIECES IN &IGURES AND BY ADDING MULTIPLEXORS 4WO MULTIPLEXORS ARE NEEDED AS DESCRIBED IN THE EXAMPLE #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 15 Branch Instructions
• Read register operands
• Compare operands (use ALU: subtract and check zero output)
• Calculate target address
• Sign-extend displacement
• Shift left two places (word displacement)
• Add to PC+4 (already calculated by instruction fetch)
CSEE 3827, Spring 2009 Martha Kim 16 Part 3: Instruction Fetch w. Branch
0# FROM INSTRUCTION DATAPATH
"RANCH 3UM !DD TARGET 3HIFT LEFT
2EAD !,5 OPERATION )NSTRUCTION REGISTER 2EAD 2EAD DATA REGISTER 4O BRANCH :ERO 2EGISTERS !,5 CONTROL LOGIC 7RITE REGISTER 2EAD DATA 7RITE DATA 2EG7RITE
3IGN EXTEND
1, Ê{°Ê / iÊ`>Ì>«>Ì ÊvÀÊ>ÊLÀ>V ÊÕÃiÃÊÌ iÊ 1ÊÌÊiÛ>Õ>ÌiÊÌ iÊLÀ>V ÊV`ÌÊ >`Ê>ÊÃi«>À>ÌiÊ>``iÀÊÌÊV«ÕÌiÊÌ iÊLÀ>V ÊÌ>À}iÌÊ>ÃÊÌ iÊÃÕÊvÊÌ iÊVÀiiÌi`Ê* Ê >`ÊÌ iÊÃ}iÝÌi`i`]ÊÜiÀÊ£ÈÊLÌÃÊvÊÌ iÊÃÌÀÕVÌÊÌ iÊLÀ>V Ê`ë>ViiÌ®]Êà vÌi`Ê ivÌÊÓÊLÌðÊ4HE UNIT LABELED 3HIFT LEFT IS SIMPLY A ROUTING OF THE SIGNALS BETWEEN INPUT AND OUTPUT THAT ADDS TWO TO THE LOW ORDER END OF THE SIGN EXTENDED OFFSET l ELD NO ACTUAL SHIFT HARDWARE IS NEEDED SINCE THE AMOUNT OF THE hSHIFTv IS CONSTANT 3INCE WE KNOW THAT THE OFFSET WAS SIGN EXTENDED FROM BITS THE SHIFT WILL THROW AWAY ONLY hSIGN BITSv #ONTROL LOGIC IS USED TO DECIDE WHETHER THE INCREMENTED 0# OR BRANCH TARGET SHOULD REPLACE THE 0# BASED ON THE :ERO OUTPUT OF THE !,5 #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 17 Full Datapath
0#3RC
- !DD U X !,5 !DD RESULT 3HIFT LEFT
2EAD 2EAD !,53RC !,5 OPERATION 0# REGISTER ADDRESS 2EAD -EM7RITE DATA 2EAD -EMTO2EG REGISTER :ERO )NSTRUCTION 2EGISTERS !,5 !,5 2EAD 2EAD !DDRESS 7RITE RESULT DATA )NSTRUCTION REGISTER DATA - - U U MEMORY X 7RITE X DATA 7RITE $ATA 2EG7RITE DATA MEMORY
-EM2EAD 3IGN EXTEND
1, Ê{°££Ê / iÊëiÊ`>Ì>«>Ì ÊvÀÊÌ iÊ *-Ê>ÀV ÌiVÌÕÀiÊVLiÃÊÌ iÊiiiÌÃÊÀiµÕÀi`ÊLÞÊ`vviÀiÌÊÃÌÀÕVÌÊ V>ÃÃiðÊ4HE COMPONENTS COME FROM &IGURES AND Ê4HIS DATAPATH CAN EXECUTE THE BASIC INSTRUCTIONS LOAD STORE WORD !,5 OPERATIONS AND BRANCHES IN A SINGLE CLOCK CYCLE !N ADDITIONAL MULTIPLEXOR IS NEEDED TO INTEGRATE BRANCHES 4HE SUPPORT FOR JUMPS WILL BE ADDED LATER #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 18 MIPS Datapath Control Datapath Control Scheme
Main control controls whole - !DD U X datapath based on opcode !,5 !DDRESULT 3HIFT 2EG$ST LEFT "RANCH -EM2EAD )NSTRUCTION ;n= -EMTO2EG #ONTROL !,5/P -EM7RITE !,53RC 2EG7RITE
)NSTRUCTION ;n= 2EAD ALU control controls ALU 2EAD REGISTER 0# ADDRESS 2EAD based on opcode )NSTRUCTION ;n= 2EAD DATA REGISTER :ERO )NSTRUCTION (ALUOp) and function ;n= !,5 !,5 2EAD - 7RITE 2EAD !DDRESS U RESULT DATA )NSTRUCTION )NSTRUCTION ;n= REGISTER DATA - - field (funct) X U U MEMORY X 7RITE X DATA 2EGISTERS 7RITE $ATA DATA MEMORY
)NSTRUCTION ;n= 3IGN !,5 EXTEND CONTROL
)NSTRUCTION ;n=
1, Ê{°£ÇÊ / iÊëiÊ`>Ì>«>Ì ÊÜÌ ÊÌ iÊVÌÀÊÕÌ°Ê4HE INPUT TO THE CONTROL UNIT IS THE BIT OPCODE l ELD FROM THE INSTRUCTION 4HE OUTPUTS OF THE CONTROL UNIT CONSIST OF THREE BIT SIGNALS THAT ARE USED TO CONTROL MULTIPLEXORS 2EG$ST !,53RC AND -EMTO2EG THREE SIGNALS FOR CON TROLLING READS AND WRITES IN THE REGISTER l LE AND DATA MEMORY 2EG7RITE -EM2EAD AND -EM7RITE A BIT SIGNAL USED IN DETERMINING WHETHER TO POSSIBLY BRANCH "RANCH AND A BIT CONTROL SIGNAL FOR THE !,5 !,5/P !N !.$ GATE IS USED TO COMBINE THE BRANCH CONTROL SIGNAL AND THE :ERO OUTPUT FROM THE !,5 THE !.$ GATE OUTPUT CONTROLS THE SELECTION OF THE NEXT 0# .OTICE THAT 0#3RC IS NOW A DERIVED SIGNAL RATHER THAN ONE COMING DIRECTLY FROM THE CONTROL UNIT 4HUS WE DROP THE SIGNAL NAME IN SUBSEQUENT l GURES #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED CSEE 3827, Spring 2009 Martha Kim 20 ALU Control Inputs/Outputs
Main Control R-type → 10 ALUOp lw → 00 2 ALU 0000 → AND sw → 00 Operation 0001 → OR beq → 01 ALU control 4 0010 → add
Instruction[5:0] 0110 → subtract
0111 → set on less than
(See Appendix C of text for implementation of corresponding ALU.)
CSEE 3827, Spring 2009 Martha Kim 21 ALU Control Implementation
opcode ALUOp from main control Instruction[5:0] Operation lw → 00 xxxxxx → load word → add → 0010
sw → 00 xxxxxx → store word → add → 0010
beq → 01 xxxxxx → branch equal → subtract → 0110
R-type → 10 100000 → add → add → 0010
R-type → 10 100010 → subtract → subtract → 0110
R-type → 10 100100 → AND → AND → 0000
R-type → 10 100101 → OR → OR → 0001
R-type → 10 101010 → set on less than → set on less than → 0111
CSEE 3827, Spring 2009 Martha Kim 22 ALU Control Truth Table
ALUOp from main control Instruction[5:0] Operation
00 xxxxxx 0010
00 xxxxxx 0010
01 xxxxxx 0110
10 100000 0010
10 100010 0110
10 100100 0000
10 100101 0001
10 101010 0111
CSEE 3827, Spring 2009 Martha Kim 23 ALU Control Truth Table 2
!,5/P &UNCT l ELD !,5/P !,5/P & & & & & & /PERATION 888888 888888 88 8 88 88 88 8 88
&)'52% 4HE TRUTH TABLE FOR THE !,5 CONTROL BITS CALLED /PERATION 4HE INPUTS ARE THE !,5/P AND FUNCTION CODE l ELD /NLY THE ENTRIES FOR WHICH THE !,5 CONTROL IS ASSERTED ARE SHOWN 3OME DONT CARE ENTRIES HAVE BEEN ADDED &OR EXAMPLE THE !,5/P DOES NOT USE THE ENCODING SO THE TRUTH TABLE CAN CONTAIN ENTRIES 8 AND 8 RATHER THAN AND .OTE THAT WHEN THE FUNCTION l ELD IS USED THE l RST BITS & AND & OF THESE INSTRUCTIONS ARE ALWAYS SO THEY ARE DONT CARE TERMS AND ARE REPLACED WITH 88 IN THE TRUTH TABLE #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 24 Datapath Control Scheme
- !DD U X !,5 !DDRESULT 3HIFT 2EG$ST LEFT "RANCH -EM2EAD )NSTRUCTION ;n= -EMTO2EG #ONTROL !,5/P -EM7RITE !,53RC 2EG7RITE
)NSTRUCTION ;n= 2EAD 2EAD REGISTER 0# ADDRESS 2EAD )NSTRUCTION ;n= 2EAD DATA REGISTER :ERO )NSTRUCTION ;n= !,5 !,5 2EAD - 7RITE 2EAD !DDRESS U RESULT DATA )NSTRUCTION )NSTRUCTION ;n= REGISTER DATA - - X U U MEMORY X 7RITE X DATA 2EGISTERS 7RITE $ATA DATA MEMORY
)NSTRUCTION ;n= 3IGN !,5 EXTEND CONTROL
)NSTRUCTION ;n=
1, Ê{°£ÇÊ / iÊëiÊ`>Ì>«>Ì ÊÜÌ ÊÌ iÊVÌÀÊÕÌ°Ê4HE INPUT TO THE CONTROL UNIT IS THE BIT OPCODE l ELD FROM THE INSTRUCTION 4HE OUTPUTS OF THE CONTROL UNIT CONSIST OF THREE BIT SIGNALS THAT ARE USED TO CONTROL MULTIPLEXORS 2EG$ST !,53RC AND -EMTO2EG THREE SIGNALS FOR CON TROLLING READS AND WRITES IN THE REGISTER l LE AND DATA MEMORY 2EG7RITE -EM2EAD AND -EM7RITE A BIT SIGNAL USED IN DETERMINING WHETHER TO POSSIBLY BRANCH "RANCH AND A BIT CONTROL SIGNAL FOR THE !,5 !,5/P !N !.$ GATE IS USED TO COMBINE THE BRANCH CONTROL SIGNAL AND THE :ERO OUTPUT FROM THE !,5 THE !.$ GATE OUTPUT CONTROLS THE SELECTION OF THE NEXT 0# .OTICE THAT 0#3RC IS NOW A DERIVED SIGNAL RATHER THAN ONE COMING DIRECTLY FROM THE CONTROL UNIT 4HUS WE DROP THE SIGNAL NAME IN SUBSEQUENT l GURES #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED CSEE 3827, Spring 2009 Martha Kim 25 Main control signals derive from instruction types
R-type: 0 rs rt rd shamt funct 31:26 25:21 20:16 15:11 10:6 5:0
Load/Store: 35 or 43 rs rt constant 31:26 25:21 20:16 15:0
Branch: 4 rs rt constant 31:26 25:21 20:16 15:0 sign-extend always read, write for and add read except R-type for load and load CSEE 3827, Spring 2009 Martha Kim 26 R-Type Control Signals
0
1
0 0 0
1
0
10 (Alt. illustration: Fig. 4.19) CSEE 3827, Spring 2009 Martha Kim27 lw Control Signals
0
1
0 1 1
0
1
00 (Alt. illustration: Fig. 4.20) CSEE 3827, Spring 2009 Martha 28Kim sw Control Signals
0
0
1 1 x
x
0
00
CSEE 3827, Spring 2009 Martha Kim 29 beq Control Signals
1
0
0 0 x
x
0
01 (Alt. illustration: Fig. 4.21) CSEE 3827, Spring 2009 Martha Kim 30 Main Control Truth Table
iÌ ,i} i i Instruction[31:26]ÃÌÀÕVÌ ,i} ÃÌ 1-ÀV ,i} 7ÀÌi ,i>` 7ÀÌi À>V 1"«£ 1"«ä 000000 ,vÀ>Ì £ ä ä£ää ä £ ä 100011 LW ä £ £££ä ä ä ä 101011 SW 8 £ 8ää£ ä ä ä 000100 BEQ 8 ä 8äää £ ä £ 1, Ê{°£nÊ / iÊÃiÌÌ}ÊvÊÌ iÊVÌÀÊiÃÊÃÊV«iÌiÞÊ`iÌiÀi`ÊLÞÊÌ iÊ«V`iÊwÊi`ÃÊvÊÌ iÊÃÌÀÕVÌ°Ê4HE l RST ROW OF THE TABLE CORRESPONDS TO THE 2 FORMAT INSTRUCTIONS ADD SUB !.$ /2 AND SLT &OR ALL THESE INSTRUCTIONS THE SOURCE REGISTER l ELDS ARE RS AND RT AND THE DESTINATION REGISTER l ELD IS RD THIS DEl NES HOW THE SIGNALS !,53RC AND 2EG$ST ARE SET &URTHERMORE AN 2 TYPE INSTRUCTION WRITES A REGISTER 2EG 7RITE BUT NEITHER READS NOR WRITES DATA MEMORY 7HEN THE "RANCH CONTROL SIGNAL IS THE 0# IS UNCONDITIONALLY REPLACED WITH 0# OTHERWISE THE 0# IS REPLACED BY THE BRANCH TARGET IF THE :ERO OUTPUT OF THE !,5 IS ALSO HIGH 4HE !,5/P l ELD FOR 2 TYPE INSTRUCTIONS IS SET TO TO INDICATE THAT THE !,5 CONTROL SHOULD BE GENERATED FROM THE FUNCT l ELD 4HE SECOND AND THIRD ROWS OF THIS TABLE GIVE THE CONTROL SIGNAL SETTINGS FOR LW AND SW 4HESE !,53RC AND !,5/P l ELDS ARE SET TO PERFORM THE ADDRESS CALCULATION 4HE -EM2EAD AND -EM7RITE ARE SET TO PERFORM THE MEMORY ACCESS &INALLY 2EG$ST AND 2EG7RITE ARE SET FOR A LOAD TO CAUSE THE RESULT TO BE STORED INTO THE RT REGISTER 4HE BRANCH INSTRUCTION IS SIMILAR TO AN 2 FORMAT OPERATION SINCE IT SENDS THE RS AND RT REGISTERS TO THE !,5 4HE !,5/P l ELD FOR BRANCH IS SET FOR A SUBTRACT !,5 CONTROL WHICH IS USED TO TEST FOR EQUALITY .OTICE THAT THE -EMTO2EG l ELD IS IRRELEVANT WHEN THE 2EG7RITE SIGNAL IS SINCE THE REGISTER IS NOT BEING WRITTEN THE VALUE OF THE DATA ON THE REGISTER DATA WRITE PORT IS NOT USED 4HUS THE ENTRY -EMTO2EG IN THE LAST TWO ROWS OF THE TABLE IS REPLACED WITH 8 FOR DONT CARE $ONT CARES CAN ALSO BE ADDED TO 2EG$ST WHEN 2EG7RITE IS 4HIS TYPE OF DONT CARE MUST BE ADDED BY THE DESIGNER SINCE IT DEPENDS ON KNOWLEDGE OF HOW THE DATAPATH WORKS #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 31 Implementing Jumps The j instruction
• Unconditional jump to instruction at label
j label
• Instruction encoded in J-type format
2 address 31:26 25:0 • Jump uses word addresses
• Update PC with concatenation of: • Top 4 bits of old PC • 26-bit jump address • 00
CSEE 3827, Spring 2009 Martha Kim 33 Implementing the jump instruction
CSEE 3827, Spring 2009 Martha Kim 34 Implementing the jump instruction -- in class soln
CSEE 3827, Spring 2009 Martha Kim 35 CPU Performance Understanding Performance
• Algorithm → number of operations executed
• Programming language, compiler, architecture → determine number of machine instructions executed per operation
• Processor and memory system → determines how fast instructions are executed
• I/O system (including OS) → determines how fast I/O operations are executed
CSEE 3827, Spring 2009 Martha Kim 37 Defining Performance
• Which airplane has the best performance?
Boeing 777 Boeing 777
Boeing 747 Boeing 747
BAC/Sud BAC/Sud Concorde Concorde Douglas Douglas DC- DC-8-50 8-50
0 100 200 300 400 500 0 2000 4000 6000 8000 10000
Passenger Capacity Cruising Range (miles)
Boeing 777 Boeing 777
Boeing 747 Boeing 747
BAC/Sud BAC/Sud Concorde Concorde Douglas Douglas DC- DC-8-50 8-50
0 500 1000 1500 0 100000 200000 300000 400000
Cruising Speed (mph) Passengers x mph
CSEE 3827, Spring 2009 Martha Kim 38 Response Time and Throughput
Response time: how long it takes to do a task, sometimes also called latency [time/work]
Throughput: total work done per unit time [work/time]
How are response time and throughput affected by. . . Replacing the processor with a faster version?
Adding more processors?
For now, we’ll focus on response time CSEE 3827, Spring 2009 Martha Kim 39 Relative Performance
Define: Performance = 1 / Execution Time
“X is n times faster than Y” → Performance X / Performance Y = Execution Time Y / Execution Time X = n
Example:
Program takes 10 s to run on machine A, 15 s on machine B Execution Time B / Execution Time A = 15 / 10 = 1.5 “A is 1.5 times faster than B”
CSEE 3827, Spring 2009 Martha Kim 40 Measuring Execution Time
Define: Elapsed Time
Total response time including all aspects (Processing, I/O, overhead, idle time)
Define: CPU Time
Time spent processing a given job (discounts I/O time, other jobs shares)
Elapsed Time > CPU Time
CSEE 3827, Spring 2009 Martha Kim 41 CPU Clocking
Operation of digital hardware governed by a constant-rate clock
Clock period
Clock Data transfer and computation
Update state Time
Clock period: duration of a clock cycle e.g., 250ps = 0.25ns
Clock frequency (rate): cycles per second e.g., 4.0GHz = 4000MHz
CSEE 3827, Spring 2009 Martha Kim 42 CPU Time
CPU Time = CPU Clock Cycles * Clock Cycle Time = CPU Clock Cycles / Clock Rate
Performance improved by: 1. Reducing number of clock cycles 2. Increasing clock rate (reducing clock period)
Hardware designer must often trade off clock rate against cycle count.
CSEE 3827, Spring 2009 Martha Kim 43 CPU Time Example
Computer A: 2GHz clock, 10s CPU time
Designing Computer B: - Aim for 6s CPU Time - Clock rate increase requires 1.2x the number of cycles
How fast must Computer B’s clock be?
Clock CyclesB 1.2×Clock CyclesA Clock RateB = = CPU TimeB 6s
Clock CyclesA = CPU Time A ×Clock RateA = 10s×2GHz = 20×109 1.2×20×109 24×109 Clock Rate = = = 4GHz B 6s 6s
CSEE 3827, Spring 2009 Martha Kim 44 Instruction Count and CPI
Instruction count Determined by program, ISA, and compiler
Average cycles per instruction (CPI) - Determined by CPU hardware - If different instructions have different CPI, can compute a weighted average based on instruction mix
Clock Cycles = Instruction Count * Cycles per Instruction
CPU Time = Instruction Count * CPI * Clock Cycle Time
= (Instruction Count * CPI) / Clock Rate
CSEE 3827, Spring 2009 Martha Kim 45 CPI Example
Computer A: cycle time = 250ps, CPI=2.0 Computer B: cycle time = 500ps, CPI=1.2
Same ISA
Which is faster, and by how much?
CPU Time Instruction Count CPI Cycle Time A = × A × A = I× 2.0× 250ps = I×500ps A is faster... CPU Time Instruction Count CPI Cycle Time B = × B × B = I×1.2×500ps = I× 600ps CPU Time I× 600ps B = = 1.2 … by this much CPU Time I 500ps A ×
CSEE 3827, Spring 2009 Martha Kim 46 Amdahl’s Law
Be aware when optimizing. . .
Taffected T improved = + T unaffected improvement factor
Example: On machine A, multiplication accounts for 80s out of 100s total CPU time.
How much improvement in multiplication performance to get 5x speedup overall?
Corollary: make the common case fast
CSEE 3827, Spring 2009 Martha Kim 47 Performance Summary
Instructions Clock cycles Seconds CPU Time = x x Program Instruction Clock cycle
Algorithm, programming language and compiler compiler affect these terms.
ISA affects all three.
Performance depends on all of these things.
CSEE 3827, Spring 2009 Martha Kim 48 Single-Cycle CPU Performance Issues
• Longest delay determines clock period
• Critical path: load instruction
• instruction memory → register file → ALU → data memory → register file
• Not feasible to vary clock period for different instructions
• We will improve performance by pipelining
CSEE 3827, Spring 2009 Martha Kim 49 Pipelining Preview: Laundry Analogy
0- !- 4IME 4ASK ORDER !
"
#
$
0- !- 4IME
4ASK ORDER !
"
#
$
1, Ê{°ÓxÊ / iÊ>Õ`ÀÞÊ>>}ÞÊvÀÊ««i}° !NN "RIAN #ATHY AND $ON EACH HAVE DIRTY CLOTHES TO BE WASHED DRIED FOLDED AND PUT AWAY 4HE WASHER DRYER hFOLDER v AND hSTORERv EACH TAKE MINUTES FOR THEIR TASK 3EQUENTIAL LAUNDRY TAKES HOURS FOR LOADS OF WASH WHILE PIPELINED LAUNDRY TAKES JUST HOURS 7E SHOW THE PIPELINE STAGE OF DIFFERENT LOADS OVER TIME BY SHOWING COPIES OF THE FOUR RESOURCES ON THIS TWO DIMENSIONAL TIME LINE BUT WE REALLY HAVE JUST ONE OF EACH RESOURCE #OPYRIGHT Ú %LSEVIER )NC !LL RIGHTS RESERVED
CSEE 3827, Spring 2009 Martha Kim 50