Lnpu/oukput Dapendence X: Anb WAR
Lnpu/oukput Dapendence X: Anb WAR
peluning
S e q v e n t b e l E x e c u t i g n
PA: P X+Y
P2: P 1
Pa:RX4P
P4: S R-
p5:T R P
PE: U=S+T
P5
wite
fter
Data Dependence
RAW Read
dependence:
X>y
x t > Y : WAR
Flow
dependece: WAW
Anb X Y:
dependeuce
Oukpue
lnpu/oukput dapendence X
>Unknowu dependence
GRAH
Draw dependence qapk s1: R1<-MCA)
S2: R z R+R4
s1: Lonol R1,A
S2 Add RaR1 S3: R<-R3
Do dopendence graph
S1 R1 <- MCA)
s1: Load R1,A
S2 R2 <-MCB)
S2 Lord Rz,B $3 RL<R1+ R2
s3 Ald R1,R2 R1
Sh MCA) <-
S4: Skore AR1
S5M(6) R2
S5: Sre B, R2
s5,si s
S3 S3 S4, s2
S $3, s2
$3 53 i ot considered
5|X Since memor
51 >S1 S2 + ,
s1 S3
dvptase of ipeluning No. of lock cyclas ve
Nen-Pipelined) 6ycles
clock Pigekind4 cc
Ald:RR; R2R2t+R3
R3-R3/R1
STAGE oF
Detng in 1 cdeck eyele PieELINE
D Decode, Ex Execuiom,
Eanple:
MA: Me.mom Access,WB : Wate Back
SIRI<RI+R2 ExTi EDEA Data of Rz
S2 R2 R2
2CCC F D ExEx MAW obbained
S3 R3 +RZ CCc ongE***|* |D Ex M* |B . cc
2 34 S * *Delay 12
uesi 1 12
S-stage figeline Proces sst:
EXE for
LoAD, ADD, sUg cCe
IF,ID, Ex E ,MEM, WB 1 clock cy ola anch excep
for MUL
Exe)
,DiV: 2. cC
lesuchonm
LoAD R3,1CR2)
DV RI, R3, R
ADD R5, RG
SUB
R7RI R8
MUL RA, RRIO
22/7/22 2
M
2 R i + Rz
SI: R3
sTFID) EXE MEM |wB 6 LO
= 1 IF |IDEXE IMEnNB
Jeck y c l
Clock ccles = |l stalls = 3
3ack
IFIDExEMEMW
52 RI
R3/R IF ID ExE ExEMEMwB
s 3 R 5 - R I t R 6
IF D ExE MEMW
S4 RT -RI~28 IF ExE MEMwg
S5 R1 -R)&RID TF D EXE MEMWS)
. of Cc
Stall
12 CC= II stall = 1
at 3 hex: Ox O2MME 48 2
OPsotaddes/umedia
Taet/hestnation
MovE 9 1 , #2DRZ Ox 3D
Coodpo 1o oolaiioo|poopeooo0o0
hexOx 023A1 O030
27 t122
tEcode Cfmcion Crdo NA
, ten represet wh KX XXXX
op Cde - o f all
am tnebc gerakors
SLL sutt ft :2dadd-esso
SLL #18 R #1| R2,2
oooo00
DIDooo00ooloo0Djo x x xXKX
hex ill contain xx ia Hhe ad
lw #8,4# )
Vaelctent conke
Dolo loool olooo| O000 Obo0 o0o0
o00
S #3, 8*)
Rugiste valwe stored in
des ta ma memory
le opcgde x21
3C# )
20
1 00 pl ooosiooo
oo00 000D 0000 I o00
Soure des ahom
OxO11 o008
XOK
g0o0oilo oo
13
4
682G|
hex:
0x O1 CF
21 7 122
o MIPS
,4A3
High-lesel n g u a g e : in saD,$ai,$a2
Mnemonicsfa,h,i
Regsker,
i+j)
f Cg+h)
rejiskers
aiables
allocated to
fai)-La2+ $a3
C$aO+
$O
7MIAS
$a1
add gto, $a0,
a l , $a3
add $t1,s
$to ,$t1
Sulb $vo,
>High-leel lnguage
j= O +1O J=k
ACo)= ij
AC4J
Co) -ACo)+ nC1)
A = ACo)- ACi)
ACo ACo-AC1
ti =122
aadi $to, $o, 12
addi $t1,$0, lo
ACo) =i
Sw to , $s1) lw B , ($s);
AC=j $t2,oG5) ;
sw b2, 46sD ore efhi cint Hh doing
i 5 4 $ s / $ 2 A[ol + aC1] =
1Sw st3
subi $b2 , js, 4 ) ; t 2 = ACo)-A C1l;
Sw *ta, ¢sSI ACo) - $t2
slsl22 ata thazards
w w
utpuk
Set: Register Renaming Techniqve
R -R2 +R3 RtR3
Ta-R4-R5
R1Ry-R5
Lw R1 R FDEx MA WB
R2-C F Ex MA WB
Lw R2,c
D Ex MA wS
Lw R4,e FD Ex MA WB
RS
Lw R5,f FDEx MA WB
A3R1+R2
tDD R3,R,R2 R6-R +RS F D Ex MA
ADD R6,R, RS F Ex MA NB
EX
Sw a ,RR d R F
Si d,RG Stl
CC
cAtSE 2 MIPS lusejon
..FDEx MA wg
R1-
R-c FDEx MA 8
Ex MA WB
F
RY-e- Ex MA wB
F D
R3R+R -
Ex MA WB
D E xMA WB
RS4 F
R6-R4+R5- F D Ex MA
.F
d R6 wB
cc 16
esb+c
tg+h Lw R5, 9
Lw R1, A
LW R6
Lw R2, tDD R7, k5,R6
I5 2x1o cP I = -33
CPIRIOS
MiDsEM-2020
20 nkt
Non-Pipeun nx 40 CkC) =
/7a
20 nx4o
23
6182
1-b ooneh predchonchngre
For Cizo;i<5 i++) IS NT
aa+L
Achh onSeyvence T
Predichon
=
TTT T NT
NT TTTTT8
Accuncy 6GY.
h a State : NT 3
Ackual Segvence TT T TT NT
NT TT
NT T
NT
T
NT T NTT
fCiz 3) it+ Predacbi oM NT T
onsieo
coubnre
i+t Accuracy XX X
NT
AS:
T T TT AccuTA
TT7
P NTT
-bbrauel Preoicie ontechngve
tecknigve
Accuray =
T T
P:NTNT T
248 22 333 A
ET
IDI Fo wO 33 2 B
4 cc 33 1 c
c 44 3
35 * 3+ 3R2* 51*
49
Atnaae no. yoks
+-1),
Test
nstuchm
e Pioeune
Exete 10 iustmchn
mo
Ec stge sane aXecution Hme .find speedp "
.3
45+04
uk
yoo
26lglz : Instmetion evel faralle Lism CILP)
Pipeline deph No. of stages
=
Speednp= CPTunpipel
1+ fipelne sball cyckes pe« iustuction CPT Pipelines
5
stages 23 / Loads nert ustucki n uses load valre
oads= 5 xo.3 1
inasas
o. =
0.|15
Jmps 0-02x 1 0.0 =
Cond aoval =
o. 25x 0 .11 * 1 -
o .o475
Totl lenglby =)
29/s22 LLP
Lop n e n n l i n g
uE>*91;
=i-3)
i=i-3) /Loop vvoU twie
/LooP n
Ciloo0 ;
cmstat;
C i =R Ci)+
Li - + cnstad -21 -3 0
Ci-
TLi-= KLi-2]
+ corstat
,emsbant (5)
301 812 21 200o
F o =2.
LD Fo, olRA),
20
SD F, OCRA) / a r [o) =7
2 2606
DA R4, Ra,#3/R1=200o+1
= 2002 2008
1
Address increments by
Lp Fo,o CR4)
R1 , R1,#8
DAD
1 so
b
a: a [i++]+
5
H/Wa=i++
19l22 e ime 3 . 5 rm 6 cc
rednces
LP rdl
219 22
12 Cc
No-P ebned
Ppened=G cc CR
=
Clock Rat
iaS. n
NO. f
2
nx12 2
Specdup CR ) T
T2 2T1
AnSmes: Ti:7
6n 6
+
-
36
L6+-1ex6 ) 30 +b
C max+ delay
6 +0 6
Non-Ppelned = 129n
12
2 n
syeead-) 30t bn
Speedp=2 = 2
91122
. Cik5)
whle Ci5)
As TTNTINT
P TTTTi
2 b
T T
NTT Stng-
T NT T AS T NT
A g S T NT T
Weak-w
TTT
P TT NT T NTr P T S S W
AccA lo/,
F,D wB 1 cc
Bxe ADD =icc SUB
MUL -3 M 6
DR2 R t RO Ex WS
I D
IF D EX Ex EX WB
Ex WB
RGR5-R1 D
+ O.095 +O. 56
O.15 x o.|
x3 to.o8 x3
2
E M
R1 R2 R3
M W
ADADD R RS F
DSUB FDE M
R6 R R
DMVL FDE M W
R R DE M W
O Rs
Ro R R11
XO
5o s (5ox 5)
hne
1.0483
o.8+ 0 : 5
O8 + 0O
A.D433 15.44
Napipelave )
Pipelan
4.25
O Suko
LD F6,3CR2) Lond2 ycle
LD F2,45(R3) FPadd2 cyles
LD F, 56(RN)
Ffwwla py 10 cyoles
MULD Fo,P2,M
sUBD Fs, F6,F2
FU 3 Londer CLD/sT), 3 ADOE 2 Mulb pliers
Execuh om wite
Rsult Tsse Compete esu lt
LD P6,3(R)
Coplete
LD F2,15(R3) 2
LD FYS6(e) 5 6
MUL Fo, F2, F - 4 19
SUBD
$, F6 5 7
5 8
ISsUe Conplete Res ult
LD F6 3 (R) 2
LD F , 4s (R3 2
LD F4, 56 (R) 3 5
MULD Fo, F2 F
DiVD FiO,Fo, FG 5 21 22
SD 2doseks -1 20
-22
Mut4 cc 1
23
LD 8 cc 3 19 3 20
T12
a 25
13 DA 28
TT
|S
Supe scala Superpipelne 26l1ol 22
C1,1) =
k+(N-1) Scalar
Superscalax
tCmy1) k +(N-m) m isSwes
m ha
ving sAme k sages
ldcal Specd up
SCm, 1) = t C1,1) n CktN-4) =mCk+ N-1)
C1) k+(N) N Ck-1)
SupeTpeluing
EAch stage divide into substages
2
FE FF261 62 LFa
3
HT L2
4 cycles MP 4 + 0-25 x 50 I6.5
MR La 0.25ns
AMAT L2
MP Lz
50 yces
MPL * ME 0. o125 ns
obnl Mi
8 1 0 0 0 nctuchions
to mISSes iu L1
20 i sses îu L2
1 cqcle
L4 h tie
L2 HT 1o cy e
L2 miss PRalty =
Iod
e mstmeLim
1.5 henyretrenes=1.0
CeI
s6ume idal
A local and qlobal miss t e o enc nel AMAT, stall eyoes pey
ind
nstuceh on and hose wiout L2 a h
Local mussate f Lz / 2 Z 5o
27
wcs ake = 4 x 5° 7 =
qlobal
o 20x00 2
Tast
lo+ o. 5 x jdo Averae Heary salls pe
W AMAT 1+(0.o4)| ( 3 . - 1 ) XI.S - 3 .
L
1+0,01x 6D
3.4
Wrhuk L2 cake
160 5
5
1 + 47 x
TInst (5-1.)x1.5
AtMAT
Mwmory
Stales pe
t erage
improvemenk uith Lz
1)
teriinana
6+ 52
(3.641)
5 veent
what is t e npace ot *cmd-lerel ache asso ciaativ1ty
Civen Mhe data below,
o s miss penalby
Assu me io Cc
dine cl mapped
=
HT in L2 fr
asspcialviky increases by
ut b e o.5 Cc
2-w seke
La tov r e c e maprel 20/
LoCal i s t e f
LMR L2 fv 2-wy SA =I07.
MP L2|Od CC
increase. o 1 inCYase
ra cically
20 Cc
AMAT2w 0 t o KjoD
l + Id7xlod = 2.1 CC
Eapa-
MM Size 3 G3 2x 23
Block Sie 8KB 2 x2 2 B
Word Size 1B
Fo Assoc ahve
Mapes
Phy Sical Mdress bs splu = 33
ber
Fot 2-w SA Mo-peig
PA l,2 32 T SN 8ieck
2x13
Ta Diveckety :