I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
14
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
5
,
p
p
.
798
~
809
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
14
.i
1
.
p
p
7
9
8
-
8
0
9
798
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Seria
l para
llel dat
a
flow
-
pipeline
d p
ro
cess
ing
architec
ture
ba
sed
a
ccelerato
r f
o
r
2D trans
form
-
qua
ntizatio
n in
video
co
der and
deco
der
Su
m
a
la
t
ha
Sh
iv
a
rudra
ia
h,
Ra
j
eswa
ri
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
mm
u
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
A
c
h
a
r
y
a
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
,
B
a
n
g
a
l
o
r
e
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ap
r
9
,
2
0
2
4
R
ev
is
ed
Oct
1
7
,
2
0
2
4
Acc
ep
ted
Oct
2
1
,
2
0
2
4
Th
e
v
i
d
e
o
c
o
d
e
r
a
n
d
d
e
c
o
d
e
r
(CO
DEC)
sta
n
d
a
rd
s
fr
o
m
M
P
EG
-
4
to
th
e
re
c
e
n
t
v
e
rsa
ti
le
v
i
d
e
o
c
o
d
e
c
(VV
C),
a
d
o
p
ted
lo
ss
y
c
o
m
p
re
ss
io
n
m
e
th
o
d
o
lo
g
ies
,
wh
ich
in
v
o
l
v
e
s
tran
sfo
rm
a
ti
o
n
,
q
u
a
n
ti
z
a
ti
o
n
a
n
d
e
n
tr
o
p
y
c
o
d
in
g
.
Th
e
g
ro
win
g
u
sa
g
e
o
f
v
i
d
e
o
d
a
ta
in
a
ll
m
e
a
n
s
o
f
c
o
m
m
u
n
ica
ti
o
n
d
e
m
a
n
d
s
m
o
re
b
a
n
d
wid
t
h
a
n
d
sto
ra
g
e
re
q
u
irem
e
n
ts.
Wh
il
e
c
o
m
p
re
ss
io
n
wit
h
re
d
u
n
d
a
n
c
y
r
e
m
o
v
a
l
b
y
tran
sfo
rm
c
o
e
fficie
n
t
c
o
d
i
n
g
,
th
e
f
o
c
a
l
p
o
in
t
is
th
e
c
r
u
c
ial
se
q
u
e
n
t
ial
d
a
ta
flo
w
a
n
d
d
a
ta
p
ro
c
e
ss
in
g
str
u
c
tu
re
s.
Ha
n
d
li
n
g
t
h
e
b
l
o
c
k
wise
d
a
ta
n
e
a
r
to
th
e
p
ro
c
e
ss
in
g
u
n
it
p
ri
o
r
a
n
d
a
fter
c
o
m
p
u
tatio
n
will
re
d
u
c
e
th
e
d
a
ta
wa
it
in
g
ti
m
e
o
f
th
e
p
r
o
c
e
ss
in
g
u
n
it
,
h
e
n
c
e
a
c
c
e
lera
ti
n
g
th
e
targ
e
ted
fu
n
c
ti
o
n
a
li
ty
.
Th
e
p
ro
p
o
se
d
se
rial
p
a
ra
ll
e
l
d
a
ta
-
flo
w
p
ip
e
li
n
e
d
p
ro
c
e
ss
in
g
a
rc
h
it
e
c
t
u
re
(S
P
DPA)
a
c
c
e
ler
a
tes
th
e
sp
e
e
d
o
f
p
ro
c
e
ss
in
g
u
n
it
b
y
o
n
c
h
i
p
d
a
ta
a
v
a
il
a
b
il
it
y
a
n
d
p
a
ra
ll
e
l
d
a
ta
a
c
c
e
ss
in
g
o
p
ti
o
n
s
a
n
d
a
lso
wi
th
t
h
e
p
ip
e
li
n
e
o
p
e
ra
ti
o
n
s
o
f
tran
sfo
rm
a
ti
o
n
,
d
a
ta
tran
s
p
o
se
a
n
d
q
u
a
n
t
iza
ti
o
n
.
Th
e
p
o
st
im
p
lem
e
n
tatio
n
re
su
lt
s
o
f
th
e
a
rc
h
it
e
c
t
u
re
targ
e
ted
to
1
6
n
m
a
n
d
2
8
n
m
field
p
ro
g
ra
m
m
a
b
le
g
a
te
a
rra
y
(
F
P
G
A)
sh
o
ws
th
a
t
th
e
re
is
a
trad
e
-
o
ff
b
e
twe
e
n
p
o
we
r
a
n
d
fre
q
u
e
n
c
y
o
f
o
p
e
ra
ti
o
n
s
fo
r
v
a
ri
o
u
s
b
lo
c
k
siz
e
s.
Th
e
d
e
sig
n
targ
e
te
d
to
1
6
n
m
wo
rk
s
f
o
r
h
ig
h
e
r
fre
q
u
e
n
c
ies
wit
h
a
n
a
v
e
ra
g
e
p
o
we
r
c
o
n
su
m
p
ti
o
n
0
.
6
4
w
a
s
c
o
m
p
a
re
d
to
2
8
n
m
F
P
G
A wh
ich
c
o
n
su
m
e
s les
s a
v
e
ra
g
e
p
o
we
r
o
f
0
.
1
5
w
.
K
ey
w
o
r
d
s
:
C
o
n
tr
ast s
en
s
itiv
ity
f
u
n
ctio
n
Dis
cr
ete
co
s
in
e
tr
an
s
f
o
r
m
s
Field
p
r
o
g
r
a
m
m
ab
le
g
ate
a
r
r
a
y
Hig
h
ef
f
icien
cy
v
id
eo
co
d
in
g
Hu
m
an
v
is
u
al
s
y
s
tem
Mo
d
u
latio
n
tr
a
n
s
f
er
f
u
n
ctio
n
Ver
s
atile
v
id
eo
co
d
in
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Su
m
alath
a
Sh
iv
ar
u
d
r
aiah
R
esear
ch
Sch
o
lar
,
Dep
ar
tm
en
t
o
f
E
lectr
o
n
ics an
d
C
o
m
m
u
n
ic
atio
n
E
n
g
in
ee
r
in
g
Ach
ar
y
a
I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
B
an
g
alo
r
e,
I
n
d
ia
E
m
ail:
s
u
m
alath
a.
d
is
h
a@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Vid
eo
co
d
er
an
d
d
e
co
d
er
(
C
ODE
C
)
s
tan
d
ar
d
s
s
tar
t
in
g
wit
h
th
e
latest
v
er
s
atile
v
id
eo
co
d
ec
(
VVC)
2
0
2
0
d
o
wn
to
an
y
p
r
ev
io
u
s
s
tan
d
ar
d
h
ig
h
e
f
f
icien
cy
v
id
e
o
co
d
in
g
(
HE
VC
)
2
0
1
3
an
d
ad
v
an
ce
d
v
id
eo
co
d
i
n
g
(
AVC)
2003
h
ad
a
g
o
al
o
f
ac
h
iev
in
g
h
ig
h
v
id
e
o
q
u
ality
,
r
ed
u
cin
g
th
e
b
an
d
wid
th
an
d
s
to
r
ag
e
r
eq
u
ir
em
en
ts
[
1
]
,
[
2
]
.
R
ed
u
cin
g
b
it
r
ate
in
th
e
o
r
d
er
o
f
3
0
%
to
5
0
%
o
v
e
r
p
r
e
v
io
u
s
s
tan
d
ar
d
s
an
d
m
ain
tain
i
n
g
h
ig
h
v
id
e
o
q
u
ality
is
p
o
s
s
ib
le
o
n
ly
b
y
u
s
in
g
s
o
p
h
is
ticated
co
d
in
g
to
o
ls
an
d
alg
o
r
ith
m
s
.
B
y
a
d
o
p
tin
g
th
e
n
ew
er
co
d
i
n
g
to
o
ls
lik
e
in
cr
ea
s
in
g
th
e
n
u
m
b
e
r
o
f
in
tr
a
an
d
in
ter
p
r
ed
ictio
n
m
o
d
es,
in
clu
d
in
g
p
r
im
ar
y
tr
an
s
f
o
r
m
at
io
n
tech
n
iq
u
es
lik
e
d
is
cr
ete
co
s
in
e
tr
an
s
f
o
r
m
(
D
C
T
)
-
ty
p
e
-
I
I
/V/VI
I
I
,
d
is
cr
ete
s
in
e
tr
an
s
f
o
r
m
(
DST)
-
ty
p
e
-
I
/
VI
I
,
s
ec
o
n
d
a
r
y
l
o
w
f
r
eq
u
e
n
cy
n
o
n
s
ep
ar
ab
le
tr
a
n
s
f
o
r
m
atio
n
(
L
FNST)
ty
p
es
with
r
ec
tan
g
u
lar
tr
a
n
s
f
o
r
m
ati
o
n
,
a
n
d
f
r
eq
u
en
cy
d
ep
en
d
e
n
t a
n
d
r
ate
d
ep
en
d
en
t
p
er
ce
p
tu
al
q
u
a
n
tizatio
n
,
i
n
cr
e
ases
th
e
co
m
p
u
tatio
n
al
c
o
m
p
l
ex
ity
[
3
]
.
I
n
o
r
d
er
to
p
er
f
o
r
m
all
th
ese
co
m
p
u
tatio
n
ally
in
ten
s
iv
e
tr
an
s
f
o
r
m
atio
n
an
d
q
u
an
tizatio
n
o
f
v
id
eo
f
r
a
m
es
ef
f
icien
tly
,
th
e
h
ar
d
war
e
ar
ch
itectu
r
e
im
p
lem
en
tatio
n
r
e
p
o
r
ted
a
p
er
f
o
r
m
a
n
c
e
g
ain
o
v
e
r
s
o
f
twar
e
o
n
l
y
s
o
lu
t
io
n
s
,
esp
ec
ially
f
o
r
a
r
ea
l tim
e
p
r
o
ce
s
s
in
g
o
n
an
e
m
b
ed
d
e
d
p
latf
o
r
m
.
As th
e
n
e
w
er
a
o
f
s
ilico
n
-
on
-
ch
ip
(
So
C
)
f
ield
p
r
o
g
r
am
m
ab
le
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
S
eria
l p
a
r
a
llel d
a
ta
flo
w
-
p
i
p
elin
ed
p
r
o
ce
s
s
in
g
a
r
ch
itectu
r
e
b
a
s
ed
a
cc
elera
to
r
… (
S
u
m
a
la
th
a
S
h
iva
r
u
d
r
a
ia
h
)
799
g
ate
ar
r
ay
(
FP
GA)
r
an
g
in
g
f
r
o
m
lo
w
en
d
t
o
h
ig
h
en
d
co
m
es
with
h
y
b
r
id
p
r
o
ce
s
s
in
g
elem
en
ts
lik
e
d
ig
ital
s
ig
n
al
p
r
o
ce
s
s
o
r
s
(
DSPs
)
,
GPUS
an
d
C
PU
wh
ich
s
u
p
p
o
r
ts
h
ar
d
war
e
an
d
s
o
f
twar
e
c
o
-
d
esig
n
,
w
h
e
r
e
an
ar
ch
itect
ca
n
p
ar
titi
o
n
th
e
co
m
p
lex
v
id
eo
C
ODE
C
’
S
im
p
lem
en
tatio
n
o
n
b
o
th
C
PU
-
wh
ich
ca
n
h
an
d
le
m
o
r
e
s
eq
u
e
n
tial
d
ata
-
f
lo
w
an
d
co
n
tr
o
l
in
ten
s
iv
e
p
ar
t
wh
ile
allo
win
g
FP
GA
to
h
an
d
le
r
ec
o
n
f
ig
u
r
a
b
le
tr
an
s
f
o
r
m
an
d
q
u
an
tizatio
n
ac
ce
ler
atio
n
task
s
[
4
]
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
O
N
T
RAN
SFO
RM
A
T
I
O
N
A
ND
Q
UANT
I
Z
A
T
I
O
N
Vid
eo
C
ODE
C
i
s
r
esp
o
n
s
ib
le
to
s
atis
f
y
all
th
e
n
ee
d
s
o
f
co
n
s
u
m
er
elec
tr
o
n
ic
r
eq
u
ir
e
m
en
ts
lik
e
d
ata
s
ec
u
r
ity
,
in
ter
n
et
b
an
d
wid
th
a
n
d
s
to
r
ag
e
wh
ich
is
ev
en
tu
all
y
p
o
s
s
ib
le
b
y
en
co
d
in
g
o
f
th
e
v
id
eo
f
r
am
es
af
ter
tr
an
s
f
o
r
m
atio
n
an
d
q
u
a
n
tizatio
n
.
T
h
e
tr
an
s
f
o
r
m
atio
n
an
d
q
u
an
tizatio
n
o
f
v
i
d
eo
f
r
am
es
in
v
o
lv
es
cr
u
cial
s
eq
u
en
tial
d
ata
f
lo
w,
d
ata
ac
c
ess
an
d
tig
h
t
d
ata
d
ep
e
n
d
en
t
p
r
o
ce
s
s
in
g
s
tr
u
ctu
r
es
.
T
h
e
H
D
v
id
eo
f
r
a
m
es
ar
e
in
itially
co
n
v
er
ted
to
b
lo
ck
s
o
f
s
ize
v
ar
y
in
g
f
r
o
m
4
×4
to
1
2
8
×1
2
8
a
n
d
th
e
n
s
u
b
m
itted
f
o
r
tr
an
s
f
o
r
m
atio
n
a
n
d
q
u
an
tizatio
n
.
T
h
is
b
lo
ck
wis
e
r
esid
u
al
p
ix
els
o
f
th
e
v
id
eo
f
r
am
e
af
ter
in
tr
a
an
d
in
ter
p
r
ed
ictio
n
s
wer
e
tr
an
s
f
o
r
m
ed
f
r
o
m
s
p
atial
to
f
r
eq
u
en
cy
d
o
m
ain
u
s
in
g
c
o
s
in
e/si
n
e
tr
an
s
f
o
r
m
atio
n
tec
h
n
iq
u
e,
in
-
o
r
d
er
to
de
-
co
r
r
elate
an
im
p
o
r
tan
t
i
n
f
o
r
m
atio
n
f
r
o
m
r
ed
u
n
d
a
n
t
with
in
th
e
b
lo
ck
.
T
r
a
n
s
f
o
r
m
atio
n
f
o
llo
wed
b
y
q
u
an
tizatio
n
,
h
el
p
s
in
d
ef
in
in
g
th
e
f
in
er
lev
els
f
o
r
en
c
o
d
in
g
t
h
e
tr
an
s
f
o
r
m
c
o
ef
f
icien
ts
an
d
r
em
o
v
in
g
p
er
ce
p
tu
a
l
r
ed
u
n
d
an
t d
ata,
h
en
ce
a
b
le
to
ac
h
iev
e
f
ir
s
t l
ev
el
o
f
d
ata
co
m
p
r
ess
io
n
.
T
h
e
well
-
k
n
o
wn
tr
an
s
f
o
r
m
atio
n
ar
ch
itectu
r
e
p
r
o
p
o
s
als
f
o
r
H
E
VC
an
d
VVC
r
ely
o
n
two
1
D
p
r
o
ce
s
s
o
r
s
to
tr
an
s
f
o
r
m
r
o
ws
an
d
co
lu
m
n
s
b
y
tak
in
g
ad
v
a
n
tag
e
o
f
s
ep
ar
ab
le
p
r
o
p
er
ties
,
co
n
n
ec
ted
th
r
o
u
g
h
a
tr
an
s
p
o
s
itio
n
m
em
o
r
y
.
R
ec
o
n
f
ig
u
r
ab
le
a
r
ch
itectu
r
e
f
o
r
HE
VC
2
D
-
DC
T
,
s
u
p
p
o
r
tin
g
b
lo
c
k
s
izes
f
r
o
m
4
×
4
to
3
2
×
3
2
was
p
r
o
p
o
s
ed
b
y
[
5
]
.
T
h
e
d
esig
n
tar
g
ets
lo
g
ical
elem
en
ts
lik
e
m
u
lti
p
lier
s
an
d
DSP
b
lo
ck
s
with
lo
ca
l
s
to
r
ag
e
m
em
o
r
y
elem
en
ts
o
n
FP
GA.
T
h
e
s
y
n
t
h
esized
d
esig
n
r
ep
o
r
te
d
in
th
e
r
esu
lt
s
u
s
tain
s
4
Kp
3
0
en
co
d
i
n
g
.
T
h
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
e
u
s
ed
in
[
6
]
–
[
8
]
is
an
ev
en
-
o
d
d
d
ec
o
m
p
o
s
itio
n
-
b
a
s
ed
tr
an
s
f
o
r
m
atio
n
b
y
s
h
if
t
an
d
ad
d
u
n
its
wh
ich
is
m
o
r
e
s
u
itab
le
f
o
r
t
h
e
r
ec
o
n
f
i
g
u
r
ab
le
FP
GA
p
latf
o
r
m
th
a
n
th
e
m
atr
ix
m
u
ltip
licatio
n
m
et
h
o
d
.
T
o
r
ed
u
ce
th
e
co
m
p
u
tatio
n
al
co
m
p
lex
ity
o
f
DC
T
-
I
I
in
HE
VC
,
Me
h
er
et
a
l.
[
6
]
p
r
o
p
o
s
ed
d
if
f
er
en
t
in
t
eg
er
ap
p
r
o
x
im
ated
ar
ch
itectu
r
es
o
f
f
o
ld
ed
,
f
u
ll
p
ar
allel
s
tr
u
ctu
r
es
with
p
r
u
n
in
g
.
T
h
e
ad
d
itio
n
al
ad
d
er
tr
ee
an
d
m
u
x
es
u
s
ed
in
th
e
d
esig
n
in
cr
ea
s
e
d
ata
p
ath
c
o
m
p
lex
ity
a
n
d
laten
cy
.
T
h
e
b
u
f
f
er
u
s
ed
b
etwe
en
1
D
an
d
2
D
tr
an
s
f
o
r
m
atio
n
f
o
r
m
atr
ix
tr
an
s
p
o
s
e
o
p
er
atio
n
r
ep
o
r
ted
in
th
e
liter
atu
r
e
is
eith
er
a
co
m
b
in
atio
n
o
f
r
eg
is
ter
ar
r
a
y
an
d
m
u
ltip
lex
er
s
[
6
]
o
r
b
ased
o
n
R
AM
s
[
7
]
.
A
u
n
if
ied
ad
ap
tiv
e
m
u
lti
p
le
tr
an
s
f
o
r
m
(
AM
T
)
ar
ch
it
ec
tu
r
e
p
er
f
o
r
m
in
g
tr
an
s
f
o
r
m
atio
n
o
f
all
s
q
u
ar
e
a
n
d
asy
m
m
etr
ic
s
ize
co
m
b
in
ati
o
n
s
f
r
o
m
4
,
8
,
1
6
,
an
d
3
2
u
s
in
g
m
u
ltip
lier
I
P c
o
r
es
an
d
DSPs
s
u
itab
le
f
o
r
VVC
wer
e
p
r
esen
ted
in
[
9
]
ca
n
r
e
n
d
er
2
K
r
eso
lu
tio
n
v
i
d
eo
c
o
d
i
n
g
at
5
0
f
p
s
,
b
u
t
th
is
d
esig
n
[
9
]
is
p
r
o
p
o
s
ed
an
d
test
ed
f
o
r
en
co
d
er
p
ath
o
n
l
y
.
I
n
[
1
0
]
,
[
1
1
]
a
g
en
er
al
m
u
ltip
lier
-
b
ased
p
ip
elin
ed
2
D
-
tr
an
s
f
o
r
m
atio
n
with
d
u
al
p
o
r
t
SR
AM
in
m
atr
ix
f
o
r
m
is
u
s
ed
as
a
tr
an
s
p
o
s
e
m
em
o
r
y
.
T
h
is
a
p
p
r
o
ac
h
o
f
tr
an
s
p
o
s
e
m
em
o
r
y
u
tili
ze
s
m
o
r
e
a
r
ea
o
n
th
e
tar
g
eted
p
latf
o
r
m
.
T
h
e
p
r
i
m
ar
y
DC
T
-
I
I
an
d
s
ec
o
n
d
ar
y
tr
an
s
f
o
r
m
s
lik
e
DC
T
-
VI
I
I
an
d
DST
-
VI
I
h
a
r
d
war
e
im
p
lem
en
tatio
n
s
[
1
2
]
–
[
1
4
]
r
ep
o
r
ted
g
o
o
d
p
er
f
o
r
m
an
ce
o
n
S
o
C
FP
GA.
T
h
e
n
ew
in
teg
er
DC
T
co
ef
f
icien
ts
d
er
iv
ed
in
[
1
5
]
h
ad
a
g
o
al
o
f
s
im
ilar
p
er
f
o
r
m
a
n
ce
co
m
p
a
r
ed
to
th
e
o
r
ig
in
al
DC
T
,
with
a
tr
ad
e
-
o
f
f
b
etwe
en
r
eso
u
r
ce
c
o
s
t
an
d
co
m
p
r
ess
io
n
.
An
a
r
ch
i
tectu
r
e
s
u
p
p
o
r
tin
g
all
tr
a
n
s
f
o
r
m
s
izes
o
f
HE
VC
b
y
r
ec
u
r
s
iv
e
p
r
o
ce
s
s
in
g
was
p
r
o
p
o
s
ed
b
y
[
1
6
]
in
v
o
lv
es
id
e
n
tify
i
n
g
th
e
n
u
m
b
er
o
f
p
ip
elin
ed
r
eg
is
ter
s
to
b
e
i
n
clu
d
ed
in
th
e
cr
itical
p
ath
to
o
b
tain
all
th
e
o
u
tp
u
ts
in
a
s
in
g
le
clo
ck
c
y
cle.
T
h
e
r
ec
e
n
t V
L
SI
im
p
lem
en
tatio
n
o
f
i
n
teg
er
ar
ch
itectu
r
e
b
ased
o
n
o
b
f
u
s
c
atio
n
tech
n
iq
u
e
an
d
s
y
s
to
lic
ar
r
ay
s
tr
u
ctu
r
e
[
1
7
]
with
m
in
im
al
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
r
ep
o
r
ted
g
o
o
d
s
p
ee
d
an
d
lo
w
p
o
wer
co
n
s
u
m
p
tio
n
.
T
h
e
u
s
u
al
co
d
i
n
g
to
o
l
ap
p
li
ed
af
ter
tr
an
s
f
o
r
m
atio
n
is
t
h
e
q
u
an
tizatio
n
to
r
em
o
v
e
p
er
ce
p
tu
al
r
ed
u
n
d
an
cy
.
Qu
an
tizatio
n
p
r
o
ce
s
s
in
v
id
eo
co
d
ec
s
tan
d
ar
d
p
lay
s
a
cr
u
cial
r
o
le
in
ac
h
ie
v
in
g
h
ig
h
co
m
p
r
ess
io
n
ef
f
icien
cy
with
o
u
t
s
ig
n
if
ican
t
l
o
s
s
in
v
is
u
al
q
u
ality
.
T
h
e
is
o
tr
o
p
ic
h
u
m
an
v
is
u
al
s
y
s
tem
(
HVS)
m
o
d
el
p
r
o
p
o
s
ed
b
y
Daly
is
ad
o
p
ted
f
o
r
DC
T
b
ased
J
PEG
im
ag
e
c
o
m
p
r
ess
io
n
to
d
er
i
v
e
a
p
er
ce
p
t
u
ally
a
d
ap
tiv
e
q
u
an
tizatio
n
tab
le
[
1
8
]
,
w
h
ich
is
u
s
ed
as
a
d
ef
au
lt
QM
in
tr
a
m
atr
i
x
in
H
E
VC
.
Kee
p
in
g
HVS
-
co
n
tr
ast
s
en
s
itiv
ity
f
u
n
ctio
n
(
C
SF
)
m
o
d
el
in
m
in
d
,
th
e
f
r
eq
u
en
cy
b
ased
q
u
an
tizatio
n
m
at
r
ix
(
QM
)
[
1
9
]
is
s
u
g
g
ested
to
s
ca
le
lo
w
f
r
eq
u
en
c
y
co
ef
f
icien
ts
b
y
f
in
er
v
alu
es
th
a
n
h
ig
h
f
r
eq
u
en
cy
with
in
th
e
tr
a
n
s
f
o
r
m
ed
b
lo
ck
.
T
h
e
d
ef
au
lt
f
r
eq
u
en
cy
d
ep
e
n
d
en
t
QM
b
ased
o
n
in
tr
a
a
n
d
in
ter
p
r
ed
icted
ty
p
e
tr
a
n
s
f
o
r
m
b
lo
ck
s
with
tr
an
s
f
o
r
m
s
ize
is
d
ef
in
ed
[
1
9
]
o
n
ly
f
o
r
4
×4
an
d
8
×
8
s
ize.
Fo
r
h
ig
h
er
b
lo
c
k
s
ize
th
e
8
×8
s
ize
m
atr
ix
v
alu
es
ar
e
r
ep
ea
te
d
o
n
e
to
2
×2
p
atter
n
f
o
r
1
6
×
1
6
s
ize
m
atr
ix
an
d
o
n
e
to
4
×4
p
atter
n
f
o
r
3
2
×
3
2
s
ize
m
atr
ix
r
esp
ec
tiv
ely
.
T
h
e
co
m
m
o
n
ly
ap
p
lied
b
lo
ck
b
ased
lo
s
s
y
v
id
eo
c
o
m
p
r
ess
io
n
t
o
o
ls
to
m
ee
t
n
etwo
r
k
r
eq
u
i
r
em
en
t
o
f
u
ltra
h
ig
h
d
ef
in
itio
n
telev
is
io
n
(
UHDT
V)
,
h
as
to
h
an
d
le
p
r
o
b
lem
s
lik
e
b
lo
ck
in
g
,
r
in
g
in
g
an
d
b
lu
r
r
i
n
g
ar
tifa
cts.
T
h
e
co
n
to
u
r
in
g
ar
tifa
cts
ar
e
m
o
s
t
co
m
m
o
n
ly
n
o
ticea
b
le
in
u
ltra
h
i
g
h
d
ef
in
i
tio
n
(
UHD)
d
is
p
lay
s
,
b
ec
au
s
e
o
f
c
o
ar
s
ely
q
u
an
tized
h
ig
h
f
r
eq
u
en
cy
v
alu
es
b
y
s
ca
lar
q
u
an
tizatio
n
.
T
o
av
o
i
d
th
is
co
n
to
u
r
i
n
g
p
r
o
b
lem
,
a
n
a
d
ap
tiv
e
q
u
an
tize
v
alu
es
to
b
e
co
n
s
id
er
ed
[
2
0
]
t
o
av
o
id
ze
r
o
in
g
o
f
d
ea
d
zo
n
e
v
a
lu
es
an
d
f
alse
ed
g
es.
T
h
e
im
p
r
o
v
em
en
t
in
HVS
-
C
SF
m
o
d
el
f
o
r
h
ig
h
r
eso
lu
tio
n
d
is
p
lay
s
s
u
g
g
ested
b
y
[
2
1
]
a
n
d
d
ev
elo
p
ed
ad
ap
tiv
e
QM
f
o
r
s
ca
lab
le
HE
VC
,
wh
er
e
h
ig
h
f
r
e
q
u
en
cy
co
ef
f
icien
ts
wer
e
q
u
an
tized
with
lo
wer
wei
g
h
t
v
alu
es
a
n
d
h
e
n
ce
h
a
d
to
p
a
y
a
b
it
m
o
r
e
b
u
d
g
et.
T
h
e
p
er
ce
p
tu
al
r
ed
u
n
d
a
n
cies
ar
e
ex
p
l
o
ited
b
y
c
o
m
b
in
i
n
g
t
h
e
lo
s
s
less
tr
an
s
f
o
r
m
s
tep
with
q
u
a
n
tizatio
n
in
all
th
e
v
id
e
o
C
ODE
C
s
tan
d
ar
d
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
5
:
798
-
8
0
9
800
So
th
e
p
o
s
itio
n
im
p
o
r
tan
ce
o
f
th
e
tr
an
s
f
o
r
m
ed
p
ix
els b
y
E
u
cl
id
ea
n
d
is
tan
ce
m
ea
s
u
r
em
en
t
f
r
o
m
DC
co
ef
f
icie
n
t
to
all
o
t
h
er
AC
co
ef
f
icien
ts
in
L
u
m
a
a
n
d
C
h
r
o
m
a
C
b
an
d
C
r
t
r
an
s
f
o
r
m
b
lo
c
k
s
an
d
n
o
r
m
alize
d
d
is
p
lay
r
eso
lu
tio
n
h
y
p
o
te
n
u
s
e
f
o
r
QM
d
e
r
iv
atio
n
is
co
n
s
id
er
ed
[
2
1
]
,
[
2
2
]
.
Fit
tin
g
th
e
co
m
p
le
x
B
ar
ten
’
s
C
SF
m
o
d
el
to
Daly
m
o
d
el,
s
u
itab
le
f
o
r
h
ig
h
d
y
n
a
m
ic
r
an
g
e
(
HDR
)
-
u
ltra
h
ig
h
d
ef
i
n
itio
n
v
id
eo
,
d
e
v
elo
p
e
d
[
2
3
]
f
o
r
b
o
th
L
u
m
a
an
d
C
h
r
o
m
a
C
SF
tu
n
ed
f
r
eq
u
en
cy
weig
h
tin
g
m
atr
ices
(
FW
M)
f
o
r
8
×8
tr
a
n
s
f
o
r
m
u
n
it
(
T
U)
s
ize.
T
h
en
th
is
m
atr
ix
ca
n
b
e
u
p
an
d
d
o
wn
s
am
p
led
to
d
e
r
iv
e
QM
f
o
r
o
th
er
s
izes.
T
h
e
f
ir
s
t
QM
f
o
r
L
u
m
a
an
d
C
h
r
o
m
a
co
d
i
n
g
was in
v
esti
g
ated
b
y
co
n
s
id
er
in
g
th
e
C
SF
o
f
DC
T
s
u
b
b
an
d
s
f
o
r
R
GB
v
id
eo
s
[
2
4
]
.
T
h
e
R
,
G
,
an
d
B
ch
a
n
n
els
wer
e
co
m
b
in
e
d
with
1
:1
:4
r
atio
s
with
h
ig
h
p
r
io
r
ity
ass
ig
n
ed
to
G
-
ch
an
n
el
f
o
r
QM
d
er
iv
atio
n
.
T
h
e
v
is
u
al
q
u
ality
m
etr
ic
an
aly
s
is
an
d
th
e
co
r
r
esp
o
n
d
in
g
ex
p
er
im
en
t
al
wo
r
k
s
ar
e
r
ev
iewe
d
d
etail
in
[
2
5
]
w
h
ich
ar
e
m
o
r
e
o
n
HVS.
T
h
ey
also
s
u
g
g
ested
lear
n
in
g
b
ased
ad
ap
tiv
e
q
u
an
ti
za
tio
n
ca
n
ad
v
an
ce
th
e
p
e
r
f
o
r
m
an
ce
an
d
will
also
b
e
s
u
itab
l
e
f
o
r
m
ac
h
in
e
v
is
io
n
ap
p
licatio
n
s
.
Fro
m
th
e
liter
atu
r
e
r
ev
iew,
it
is
r
ev
ea
led
th
at,
m
o
s
t
r
esear
ch
er
s
s
u
g
g
ested
d
if
f
er
e
n
t
ar
ch
it
ec
tu
r
es
f
o
r
2
D
tr
an
s
f
o
r
m
atio
n
o
f
s
q
u
ar
e
an
d
r
ec
ta
n
g
u
lar
b
lo
ck
s
izes.
Als
o
p
r
o
p
o
s
ed
i
n
teg
er
ap
p
r
o
x
im
atio
n
o
f
tr
an
s
f
o
r
m
atio
n
k
er
n
el
c
o
ef
f
ici
en
ts
s
u
itab
le
f
o
r
VL
SI
im
p
le
m
e
n
tatio
n
s
an
d
to
r
ed
u
ce
co
m
p
lex
ity
o
f
h
an
d
lin
g
r
ea
l
v
alu
es.
Nex
t
d
ef
in
ed
v
a
r
io
u
s
m
eth
o
d
s
f
o
r
tr
a
n
s
p
o
s
in
g
t
h
e
in
ter
m
ed
iate
1
D
DC
T
r
esu
lt
s
u
itab
le
f
o
r
2
D
tr
an
s
f
o
r
m
atio
n
.
T
h
en
ap
p
ly
i
n
g
q
u
a
n
tizatio
n
f
o
r
p
er
ce
p
tu
al
r
e
d
u
n
d
a
n
cy
r
em
o
v
al
a
n
d
to
s
u
p
p
o
r
t
en
t
r
o
p
y
c
o
d
in
g
.
I
n
all
th
is
wo
r
k
,
h
an
d
lin
g
m
u
ltip
le
d
ata
v
alu
es
f
r
o
m
th
e
ex
ter
n
al
s
o
u
r
ce
o
r
s
to
r
ag
e
to
p
r
o
ce
s
s
in
g
u
n
it
an
d
p
r
o
ce
s
s
in
g
d
ata
in
p
ar
allel
with
in
a
p
ip
elin
e
ar
ch
itectu
r
e
s
till
r
em
ain
ed
ch
allen
g
in
g
.
Als
o
q
u
an
tizin
g
tr
an
s
f
o
r
m
co
ef
f
icien
ts
to
g
iv
e
a
b
etter
b
al
an
ce
b
etwe
en
r
ed
u
n
d
a
n
cy
r
em
o
v
al
an
d
im
p
r
o
v
i
n
g
v
is
u
al
q
u
al
ity
o
n
a
HD
d
is
p
lay
is
a
d
o
m
ain
o
f
r
esear
ch
in
ter
est.
T
h
e
m
ain
co
n
tr
i
b
u
tio
n
o
f
th
e
p
r
o
p
o
s
ed
wo
r
k
to
ad
d
r
ess
th
e
g
ap
s
in
clu
d
e
i)
t
h
e
s
ize
o
f
d
ata
s
elec
tio
n
b
ased
o
n
b
lo
ck
s
ize
an
d
im
ag
e
wid
th
to
p
er
f
o
r
m
tr
an
s
f
o
r
m
atio
n
o
p
er
at
io
n
o
f
m
u
ltip
le
d
at
a
in
p
ar
allel.
Als
o
,
th
is
ap
p
r
o
ac
h
allo
ws f
lex
ib
le
tr
an
s
f
o
r
m
atio
n
o
f
s
q
u
ar
e
a
n
d
r
ec
ta
n
g
u
lar
d
a
ta
s
izes sp
ec
if
ied
in
n
ew
C
ODE
C
s
tan
d
ar
d
s
;
ii)
h
a
v
in
g
d
ata
n
ea
r
to
t
h
e
p
r
o
ce
s
s
in
g
u
n
it lo
ad
e
d
t
o
lin
e
b
u
f
f
er
s
an
d
n
ew
tech
n
iq
u
e
o
f
1D
-
DC
T
r
esu
lt
tr
an
s
p
o
s
e
u
s
in
g
d
em
u
x
an
d
lin
e
b
u
f
f
e
r
s
ac
ce
ler
ates
th
e
p
r
o
ce
s
s
in
g
o
f
t
r
an
s
f
o
r
m
atio
n
an
d
q
u
an
tizatio
n
o
p
er
atio
n
s
;
an
d
ii
i)
a
d
ap
tiv
e
q
u
an
tizatio
n
m
et
h
o
d
b
ased
o
n
d
is
p
lay
r
eso
lu
tio
n
to
h
av
e
a
tr
a
d
e
-
o
f
f
b
etwe
en
v
is
u
al
q
u
ality
a
n
d
n
u
m
b
er
o
f
en
co
d
in
g
b
its
p
er
p
ix
e
l d
u
r
in
g
en
tr
o
p
y
c
o
d
in
g
.
T
h
e
r
est
o
f
th
e
s
ec
tio
n
is
o
r
g
a
n
ized
as
f
o
llo
ws.
Sectio
n
3
g
iv
es
th
e
o
u
tlin
e
o
f
o
u
r
p
r
o
p
o
s
e
d
m
eth
o
d
an
d
d
etailed
m
ath
em
atica
l
m
o
d
el
with
a
n
ar
c
h
itectu
r
al
f
r
am
ewo
r
k
o
f
in
te
g
er
a
p
p
r
o
x
im
ated
2
D
-
DC
T
t
r
an
s
f
o
r
m
atio
n
an
d
p
er
ce
p
tu
ally
o
p
tim
ized
ad
ap
tiv
e
q
u
a
n
tizatio
n
m
o
d
u
les.
I
n
s
ec
tio
n
4
,
th
e
d
ata
f
lo
w
ac
ce
ler
atio
n
s
an
d
p
ip
elin
e
o
p
er
atio
n
s
o
f
p
r
o
p
o
s
ed
ar
c
h
itectu
r
e
ar
e
d
escr
ib
e
d
with
s
im
u
latio
n
an
d
i
m
p
l
e
m
e
n
t
a
ti
o
n
r
e
s
u
lt
s
o
n
t
h
e
ta
r
g
e
t
e
d
F
PG
A
e
v
al
u
a
t
i
o
n
b
o
a
r
d
s
.
F
i
n
al
l
y
,
t
h
e
c
o
n
c
l
u
s
i
o
n
is
co
v
e
r
e
d
i
n
s
e
c
t
i
o
n
5.
3.
P
RO
P
O
SE
D
M
E
T
H
O
DO
L
O
G
Y
T
h
e
ar
ch
itectu
r
al
f
r
am
ewo
r
k
i
n
Fig
u
r
e
1
g
iv
es
th
e
o
u
tlin
e
o
f
th
e
p
r
o
ce
s
s
in
g
s
u
b
-
m
o
d
u
le
s
an
d
d
ata
f
lo
w
ac
ce
ler
atio
n
.
T
h
e
test
in
p
u
t
will
b
e
l
o
ad
ed
s
er
ially
in
t
o
th
e
o
n
-
c
h
ip
lin
e
-
b
u
f
f
er
s
f
o
r
1
D
tr
an
s
f
o
r
m
atio
n
alo
n
g
th
e
r
o
w.
T
h
e
n
u
m
b
er
o
f
lin
e
b
u
f
f
er
s
in
s
tan
t
iated
will
b
e
b
ased
o
n
tr
an
s
f
o
r
m
s
ize
s
elec
tio
n
an
d
d
ep
th
o
f
ea
ch
lin
e
b
u
f
f
er
is
eq
u
al
to
im
ag
e
wid
th
s
h
o
wn
i
n
Fig
u
r
e
1
.
F
i
g
u
r
e
1
.
P
r
o
p
o
s
e
d
s
e
r
i
a
l
p
a
r
al
le
l
d
a
t
a
-
f
l
o
w
p
i
p
el
i
n
e
d
p
r
o
c
e
s
s
in
g
a
r
c
h
i
t
e
c
t
u
r
e
-
h
a
r
d
w
a
r
e
a
cc
e
le
r
a
t
o
r
f
r
a
m
e
w
o
r
k
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
S
eria
l p
a
r
a
llel d
a
ta
flo
w
-
p
i
p
elin
ed
p
r
o
ce
s
s
in
g
a
r
ch
itectu
r
e
b
a
s
ed
a
cc
elera
to
r
… (
S
u
m
a
la
th
a
S
h
iva
r
u
d
r
a
ia
h
)
801
T
h
e
m
ain
o
b
jectiv
e
o
f
r
eq
u
ir
ed
d
ata
av
ailab
ilit
y
n
ea
r
th
e
p
r
o
ce
s
s
in
g
u
n
it
is
ac
h
iev
ed
b
y
th
ese
lin
e
b
u
f
f
er
s
.
T
h
e
tr
an
s
f
o
r
m
k
er
n
el
co
ef
f
icien
ts
-
T
ar
e
lo
ad
e
d
i
n
to
a
r
ea
d
o
n
ly
m
em
o
r
y
(
R
OM
)
b
ased
o
n
tr
a
n
s
f
o
r
m
s
ize
s
elec
tio
n
.
T
h
e
f
ir
s
t
1
D
-
tr
a
n
s
f
o
r
m
atio
n
alo
n
g
r
o
w
-
wis
e
an
d
s
ec
o
n
d
2
D
-
t
r
an
s
f
o
r
m
ati
o
n
alo
n
g
c
o
lu
m
n
-
wis
e
will b
e
p
e
r
f
o
r
m
ed
b
y
g
en
e
r
al
m
atr
ix
m
u
l
tip
licatio
n
m
eth
o
d
.
Af
ter
ev
er
y
m
u
ltip
licatio
n
u
s
u
al
d
ata
tr
u
n
ca
tio
n
is
p
er
f
o
r
m
ed
b
y
s
ca
lin
g
to
m
ain
t
ain
d
ata
s
ize
o
f
1
6
-
b
it
af
ter
ev
e
r
y
s
tep
.
T
h
e
s
ca
lin
g
f
ac
to
r
d
ep
en
d
s
o
n
p
ix
el
d
ep
t
h
‘
B
’
an
d
‘
K’
wh
ich
is
a
lo
g
ar
ith
m
ic
v
alu
e
o
f
th
e
s
elec
ted
b
lo
ck
s
ize
’
N’
.
T
h
e
1
D
-
tr
an
s
f
o
r
m
atio
n
o
u
tp
u
t
will
b
e
tr
an
s
p
o
s
ed
b
y
DE
MU
X
an
d
l
in
e
b
u
f
f
e
r
s
,
wh
ich
will
b
e
f
e
d
to
th
e
s
ec
o
n
d
co
lu
m
n
-
wis
e
tr
an
s
f
o
r
m
atio
n
b
y
m
u
ltip
licatio
n
with
T
’
.
T
h
is
way
o
f
tr
an
s
f
o
r
m
atio
n
s
u
p
p
o
r
ts
b
o
th
s
q
u
ar
e
a
n
d
r
ec
tan
g
u
lar
tr
an
s
f
o
r
m
atio
n
o
f
v
ar
io
u
s
s
iz
es
as
s
h
o
wn
in
Fig
u
r
e
1
.
T
h
e
2
D
-
t
r
an
s
f
o
r
m
atio
n
o
u
tp
u
t
will
b
e
th
e
n
q
u
an
tized
b
a
s
ed
o
n
q
u
an
tizatio
n
p
ar
am
eter
,
b
lo
c
k
s
ize,
an
d
f
r
a
m
e
s
ize
co
n
f
ig
u
r
ed
b
y
co
n
tr
o
l
u
n
it
to
o
b
tain
tr
an
s
f
o
r
m
co
e
f
f
icien
t
lev
el.
As
th
e
p
r
o
p
o
s
ed
ar
c
h
itectu
r
e
is
a
u
n
if
i
ed
s
tr
u
ctu
r
e,
th
e
p
r
o
ce
s
s
o
f
f
o
r
war
d
tr
an
s
f
o
r
m
atio
n
is
p
er
f
o
r
m
ed
in
r
ev
e
r
s
e
o
r
d
er
f
o
r
in
v
e
r
s
e
tr
an
s
f
o
r
m
an
d
in
v
er
s
e
q
u
an
tizatio
n
.
T
h
e
s
er
ial
a
n
d
p
ar
allel
d
ataf
lo
w
th
r
o
u
g
h
s
u
b
m
o
d
u
les
o
f
th
e
ar
ch
itectu
r
e
with
f
o
u
r
s
tag
e
p
i
p
elin
ed
o
p
er
atio
n
s
ar
e
d
escr
ib
ed
in
d
e
tail in
th
e
n
ex
t sectio
n
s
.
3
.
1
.
I
nte
g
er
a
pp
ro
x
im
a
t
ed
2
D
-
t
ra
ns
f
o
rm
a
rc
hite
ct
ure
T
h
e
m
o
s
t
wid
ely
u
s
ed
tr
an
s
f
o
r
m
atio
n
ty
p
e
in
im
a
g
e
p
r
o
ce
s
s
in
g
an
d
v
id
eo
c
o
m
p
r
ess
io
n
s
tan
d
ar
d
is
th
e
DC
T
-
I
I
.
T
h
e
u
n
if
ied
s
ep
ar
a
b
le
2
D
tr
an
s
f
o
r
m
o
f
a
n
in
p
u
t im
a
g
e/f
r
am
e
o
f
s
ize
Mx
N
is
co
m
p
u
ted
in
(
1
)
.
T
i
,
j
=
{
1
√
N
if
i
=
0
√
2
N
c
os
(
2j
+
1
)
i
π
2N
if
i
>
0
}
(
1
)
I
n
(
1
)
,
T
i
,
j
r
e
p
r
es
en
ts
t
h
e
el
em
e
n
t
v
al
u
e
o
f
t
h
e
t
r
a
n
s
f
o
r
m
ati
o
n
co
e
f
f
ici
en
t
m
at
r
i
x
in
r
ea
l;
i
is
t
h
e
r
o
w
in
d
ex
,
j
is
t
h
e
co
l
u
m
n
i
n
d
e
x
,
N
is
t
h
e
t
r
a
n
s
f
o
r
m
s
i
ze
;
an
d
i
,
j
=
0
,
1
,
.
.
.
,
N
−
1
.
I
n
o
r
d
er
t
o
p
r
o
c
ess
t
h
e
t
r
an
s
f
o
r
m
a
ti
o
n
i
n
i
n
te
g
er
,
th
e
i
n
t
eg
e
r
ap
p
r
o
x
i
m
ati
o
n
o
f
r
ea
l
c
o
e
f
f
ic
ie
n
ts
i
n
(
1
)
c
an
b
e
o
b
ta
in
e
d
b
y
(
2
)
d
e
f
i
n
e
d
as:
,
=
[
2
∗
,
]
(
2
)
wh
er
e
n
=
6
+
2
2
=
6
+
2
an
d
N=
tr
an
s
f
o
r
m
ati
o
n
s
ize.
T
h
e
two
-
d
im
e
n
s
io
n
al
tr
an
s
f
o
r
m
o
f
an
Mx
N
b
lo
c
k
o
f
r
esid
u
al
m
atr
ix
ca
n
b
e
ac
h
ie
v
ed
b
y
(
3
)
,
f
ir
s
t
ap
p
ly
in
g
a
o
n
e
-
d
im
en
s
io
n
al
t
r
an
s
f
o
r
m
to
ea
c
h
r
o
w
o
f
th
e
b
l
o
ck
an
d
th
en
a
p
p
ly
in
g
an
o
th
e
r
o
n
e
-
d
im
en
s
io
n
al
tr
an
s
f
o
r
m
to
ea
ch
co
l
u
m
n
o
f
t
h
e
r
o
w
-
tr
a
n
s
f
o
r
m
ed
r
esu
lt.
=
′
(
3
)
T
h
e
m
atr
ix
f
o
r
m
o
f
th
e
4
-
p
o
in
t o
n
e
-
d
im
e
n
s
io
n
al
tr
an
s
f
o
r
m
is
g
iv
en
b
y
(
4
)
.
=
∗
=
[
00
01
10
11
02
03
12
13
20
21
30
31
22
23
32
33
]
=
[
00
01
10
11
02
03
12
13
20
21
30
31
22
23
32
33
]
∗
[
00
01
10
11
02
03
12
13
20
21
30
31
22
23
32
33
]
(
4
)
A
s
s
h
o
wn
in
(
4
)
,
X
r
ep
r
esen
ts
th
e
p
ix
el
r
esid
u
al
m
atr
ix
,
T
is
th
e
4
-
p
o
in
t
tr
an
s
f
o
r
m
k
e
r
n
el
m
atr
ix
,
an
d
P
is
th
e
1D
-
tr
an
s
f
o
r
m
e
d
r
esu
ltan
t
m
a
tr
ix
.
T
h
e
ca
lcu
latio
n
f
o
r
m
u
la
f
o
r
th
e
s
ec
o
n
d
lev
el
o
f
tr
a
n
s
f
o
r
m
atio
n
to
g
et
2D
-
tr
an
s
f
o
r
m
o
u
tp
u
t c
an
b
e
ex
p
r
ess
ed
as in
(
5
)
.
=
∗
′
=
[
00
01
10
11
02
03
12
13
20
21
30
31
22
23
32
33
]
=
[
00
01
10
11
02
03
12
13
20
21
30
31
22
23
32
33
]
∗
[
00
10
01
11
20
30
21
31
02
12
03
13
22
32
23
33
]
(
5
)
B
ased
o
n
th
e
ab
o
v
e
s
tep
s
o
f
co
m
p
u
tatio
n
th
e
ar
ch
itectu
r
e
is
d
esig
n
ed
as
s
h
o
wn
in
Fig
u
r
e
2
.
T
h
e
co
m
p
lete
p
ip
elin
e
ar
ch
itectu
r
e
h
as
m
ain
ly
f
o
u
r
im
p
o
r
tan
t
m
o
d
u
les:
i
)
in
p
u
t
p
ix
el
co
n
tr
o
l
u
n
it,
ii)
lin
e
-
b
u
f
f
er
s
,
iii)
p
r
o
ce
s
s
in
g
u
n
it,
an
d
iv
)
o
u
tp
u
t
b
u
f
f
er
.
T
h
e
tr
a
n
s
f
o
r
m
a
cc
eler
ato
r
m
o
d
u
le
h
as
a
v
er
y
r
eg
u
lar
s
tr
u
ctu
r
e
f
o
r
b
o
th
1
D
a
n
d
2
D
f
o
r
war
d
/in
v
er
s
e
tr
an
s
f
o
r
m
.
Hen
ce
g
iv
es a
m
o
r
e
ef
f
icien
t
p
ip
elin
in
g
o
f
th
e
s
u
b
m
o
d
u
le
as we
ll
as m
ax
im
u
m
f
r
e
q
u
en
c
y
o
f
o
p
e
r
atio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
5
:
798
-
8
0
9
802
Fig
u
r
e
2
.
Pip
elin
e
a
r
ch
itectu
r
e
o
f
2
D
-
tr
a
n
s
f
o
r
m
atio
n
3
.
1
.
1
.
I
np
ut
pix
el
co
ntr
o
l uni
t
T
h
e
m
ain
f
u
n
ctio
n
o
f
th
is
co
n
tr
o
l
u
n
it
is
to
h
an
d
le
th
e
d
a
ta
m
o
v
em
e
n
t
f
r
o
m
ex
ter
n
al
s
o
u
r
ce
s
to
lin
e
-
b
u
f
f
e
r
s
i.e
.
d
ata
wr
itin
g
o
p
er
atio
n
s
,
a
n
d
f
ee
d
to
p
r
o
ce
s
s
in
g
elem
en
ts
f
r
o
m
lin
e
-
b
u
f
f
er
s
i.e
.
d
ata
r
ea
d
in
g
o
p
er
atio
n
.
T
h
e
n
u
m
b
er
o
f
r
o
w
s
to
b
e
ac
ce
s
s
ed
f
r
o
m
th
e
in
p
u
t
f
r
am
e
d
ep
en
d
s
o
n
th
e
b
lo
c
k
s
ize
co
n
s
id
er
ed
f
o
r
o
p
er
atio
n
.
T
h
e
b
l
o
ck
s
ize
s
elec
tio
n
f
o
r
t
r
an
s
f
o
r
m
atio
n
a
n
d
q
u
an
tizatio
n
is
p
ar
am
eter
ized
,
s
o
th
e
co
r
r
esp
o
n
d
in
g
n
u
m
b
er
o
f
lin
e
-
b
u
f
f
er
s
an
d
o
th
er
f
u
n
ctio
n
al
u
n
its
ca
n
b
e
in
s
tan
tiated
ac
co
r
d
in
g
l
y
.
T
h
e
s
tate
d
iag
r
am
in
Fig
u
r
e
3
illu
s
tr
ates
th
e
r
ea
d
a
n
d
wr
ite
co
n
tr
o
l
o
f
lin
e
-
b
u
f
f
e
r
s
.
T
h
e
d
ep
th
o
f
th
e
lin
e
-
b
u
f
f
er
is
d
ec
id
ed
b
y
t
h
e
n
u
m
b
er
o
f
c
o
lu
m
n
s
in
t
h
e
in
p
u
t f
r
am
e
a
n
d
s
ize
o
f
ea
c
h
lo
ca
t
io
n
is
eq
u
al
to
in
p
u
t p
i
x
el
s
ize.
Fig
u
r
e
3
.
Fin
ite
s
tate
m
ac
h
in
e
f
o
r
in
p
u
t c
o
n
tr
o
l u
n
it
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
S
eria
l p
a
r
a
llel d
a
ta
flo
w
-
p
i
p
elin
ed
p
r
o
ce
s
s
in
g
a
r
ch
itectu
r
e
b
a
s
ed
a
cc
elera
to
r
… (
S
u
m
a
la
th
a
S
h
iva
r
u
d
r
a
ia
h
)
803
3
.
1
.
2
.
L
ine
-
bu
f
f
er
s
As
s
h
o
wn
in
th
e
h
ar
d
war
e
a
r
c
h
itectu
r
e
f
r
a
m
ewo
r
k
in
Fig
u
r
e
2
,
t
h
e
lin
e
-
b
u
f
f
e
r
s
ar
e
u
s
ed
at
th
e
in
p
u
t
s
tag
e
o
f
1
D
an
d
2
D
tr
an
s
f
o
r
m
s
.
T
h
e
t
o
tal
d
ep
t
h
o
f
th
e
li
n
e
-
b
u
f
f
er
s
u
s
ed
at
th
e
in
p
u
t
o
f
b
o
th
1
D
an
d
2
D
tr
an
s
f
o
r
m
s
is
eq
u
al
to
th
e
co
lu
m
n
v
alu
e
o
f
th
e
in
p
u
t
f
r
am
e
.
T
h
e
d
ata
wid
th
o
f
lin
e
-
b
u
f
f
er
u
s
ed
b
ef
o
r
e
1
D
tr
an
s
f
o
r
m
atio
n
is
8
-
b
it,
as
it
s
to
r
es
in
p
u
t
p
ix
el
v
alu
es
an
d
th
e
d
ata
wid
th
o
f
lin
e
-
b
u
f
f
er
u
s
ed
at
th
e
in
p
u
t
o
f
2
D
tr
an
s
f
o
r
m
atio
n
is
a
1
D
-
DC
T
o
u
tp
u
t
o
f
1
6
-
b
it.
T
h
e
r
eg
is
ter
tr
an
s
f
er
lev
el
(
R
T
L
)
elab
o
r
ated
lin
e
b
u
f
f
er
m
o
d
u
le
is
s
h
o
wn
in
Fig
u
r
e
4
.
Fig
u
r
e
4
.
R
T
L
elab
o
r
ated
lin
e
-
b
u
f
f
er
m
o
d
u
le
b
ef
o
r
e
1
D
-
t
r
an
s
f
o
r
m
atio
n
3
.
1
.
3
.
P
ro
ce
s
s
ing
un
it
I
f
th
e
b
lo
ck
s
ize
s
elec
ted
f
o
r
tr
an
s
f
o
r
m
atio
n
is
4
×
4
,
th
en
b
a
s
ed
o
n
(
4
)
,
th
e
f
o
u
r
r
o
w
elem
en
ts
o
f
th
e
tr
an
s
f
o
r
m
m
at
r
ix
co
e
f
f
icien
ts
an
d
f
o
u
r
c
o
lu
m
n
elem
en
ts
o
f
t
h
e
d
ata
m
atr
i
x
ar
e
f
ed
t
o
th
e
c
o
m
p
u
tin
g
u
n
it.
T
h
is
is
b
asically
an
elem
en
t
wis
e
m
u
ltip
lier
a
n
d
ad
d
er
.
Du
r
in
g
th
e
f
ir
s
t
1
D
tr
an
s
f
o
r
m
ati
o
n
s
tep
,
f
ir
s
t
f
o
u
r
tr
an
s
f
o
r
m
atio
n
c
o
ef
f
icien
ts
g
et
m
u
ltip
lied
with
th
e
f
o
u
r
co
l
u
m
n
d
ata
elem
en
ts
,
wh
ich
ar
e
co
n
tin
u
o
u
s
ly
s
tr
ea
m
ed
b
y
in
p
u
t
lin
e
-
b
u
f
f
e
r
s
,
th
er
e
b
y
r
esu
ltin
g
in
th
e
tr
an
s
f
o
r
m
ed
r
o
w
-
0
o
u
tp
u
t.
T
h
e
n
th
e
n
ex
t
r
o
w
f
o
u
r
tr
a
n
s
f
o
r
m
co
ef
f
icien
ts
ar
e
s
elec
ted
an
d
m
u
ltip
lied
with
th
e
p
r
ev
io
u
s
l
y
s
elec
ted
co
lu
m
n
d
ata
v
alu
e
s
to
g
en
er
ate
r
o
w
-
1
o
u
tp
u
t
v
alu
es
o
f
a
1
D
-
tr
an
s
f
o
r
m
atio
n
.
Fo
r
d
ec
id
in
g
th
e
1
D
-
r
o
w
wis
e
tr
an
s
f
o
r
m
atio
n
o
p
e
r
a
tio
n
th
e
4
:1
m
u
x
is
u
s
ed
.
T
h
e
o
b
tain
ed
1
D
-
tr
a
n
s
f
o
r
m
ed
v
alu
es
ar
e
tr
a
n
s
p
o
s
ed
u
s
in
g
1
:4
d
e
m
u
x
a
n
d
th
e
s
ec
o
n
d
s
et
o
f
lin
e
-
b
u
f
f
er
s
.
T
h
ese
1
D
-
tr
an
s
f
o
r
m
ed
v
al
u
es
ar
e
ap
p
lied
t
o
th
e
co
l
u
m
n
-
wis
e
2
D
tr
an
s
f
o
r
m
atio
n
m
o
d
u
le
as
d
ep
icted
in
(
5
)
.
T
h
e
r
esu
lt o
f
th
is
m
o
d
u
le
is
th
e
r
o
w
-
wis
e
f
in
al
o
u
tp
u
t o
f
th
e
tr
a
n
s
f
o
r
m
ed
f
r
am
e.
3
.
1
.
4
.
O
utput
bu
f
f
er
T
h
e
tr
an
s
f
o
r
m
ed
r
esu
lts
ar
e
s
en
t
s
eq
u
en
tially
to
t
h
is
o
u
tp
u
t
b
u
f
f
er
,
w
h
ich
is
b
asically
an
A
XI
-
s
tr
ea
m
b
ased
s
in
g
le
clo
ck
en
ab
led
FI
FO.
T
h
e
d
ep
t
h
o
f
FIFO
is
k
ep
t
1
6
b
y
tes
an
d
wid
th
is
1
6
-
b
it.
T
h
r
o
u
g
h
th
is
b
u
f
f
e
r
m
o
d
u
le,
r
esu
ltan
t
d
ata
ca
n
b
e
s
en
t
to
th
e
DR
AM
m
em
o
r
y
an
d
to
th
e
q
u
a
n
tizatio
n
m
o
d
u
le
o
r
to
an
y
tr
a
n
s
f
o
r
m
d
ep
en
d
e
n
t p
r
o
ce
s
s
in
g
m
o
d
u
le,
i.e
.
ev
en
f
o
r
i
n
v
er
s
e
tr
an
s
f
o
r
m
atio
n
wit
h
o
u
t
q
u
an
tizatio
n
.
3
.
2
.
P
er
ce
ptua
lly
a
da
ptiv
e
f
re
qu
ency
depe
nd
ent
q
ua
ntiz
a
t
io
n
T
h
e
ar
ch
itectu
r
e
m
ain
ly
co
n
s
i
d
er
s
th
e
ad
a
p
tiv
e
QM
d
e
r
iv
ed
b
ased
o
n
d
is
p
lay
r
eso
lu
tio
n
p
ar
am
eter
‘
w’
,
th
e
p
er
ce
p
tu
al
im
p
o
r
ta
n
t
weig
h
t
v
alu
e
c
o
n
s
id
er
atio
n
o
n
d
is
tan
ce
m
ea
s
u
r
em
en
t
‘
E
u
d
’
b
etwe
en
DC
an
d
AC
co
ef
f
icien
ts
with
in
TU
b
l
o
c
k
s
ize
an
d
f
in
ally
m
o
d
if
y
i
n
g
2
D
FW
M,
H(
u
,
v
)
.
I
n
th
is
en
tire
p
r
o
ce
s
s
th
e
m
eth
o
d
o
l
o
g
y
p
r
o
p
o
s
ed
in
[
21
]
,
[
22
]
is
f
o
llo
wed
an
d
th
e
p
ar
am
eter
s
lis
ted
wer
e
m
o
d
if
ied
in
eq
u
atio
n
s
,
ex
p
lain
e
d
in
d
etail
as f
o
llo
ws
.
3
.
2
.
1
.
Q
ua
ntiz
a
t
io
n
m
a
t
rix
ba
s
ed
o
n dis
p
la
y
re
s
o
lutio
n
B
ased
o
n
d
is
p
lay
s
cr
ee
n
s
ize,
t
h
e
n
o
r
m
alize
d
h
y
p
o
ten
u
s
e
v
alu
e
p
ar
am
eter
‘
h
’
,
in
p
ix
els
ca
n
b
e
m
o
d
ele
d
as
in
(
6
)
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
5
:
798
-
8
0
9
804
ℎ
=
ℎ
ℎ
ℎ
∈
[
0
,
1
]
(
6
)
wh
er
e,
ℎ
ℎ
=
th
eo
r
etica
l
m
ax
im
u
m
h
y
p
o
ten
u
s
e
v
alu
e
,
in
p
i
x
els
an
d
ℎ
=
ac
tu
al
m
ax
im
u
m
h
y
p
o
te
n
u
s
e
v
alu
e,
i
n
p
ix
els.
C
alcu
latin
g
‘
h
’
b
ased
o
n
r
eso
l
u
tio
n
o
f
d
is
p
lay
u
n
it,
d
ep
en
d
s
o
n
h
theoretic
al
an
d
h
actual
d
ef
i
n
ed
b
y
u
s
in
g
(
7
)
an
d
(
8
)
:
ℎ
ℎ
=
√
2
+
2
(
7
)
ℎ
=
√
2
+
2
(
8
)
T
h
e
th
eo
r
etica
l
m
ax
im
u
m
p
i
x
el
v
alu
es
an
d
o
n
th
e
m
a
x
im
u
m
p
o
s
s
ib
le
im
ag
e
s
ize,
in
p
ix
els,
p
er
m
itted
in
th
e
J
PEG
s
tan
d
ar
d
is
6
5
5
3
5
×
6
5
5
3
5
[
1
4
]
.
T
h
er
e
f
o
r
e,
s
u
b
s
titu
tin
g
=6
5
5
3
5
an
d
=6
5
5
3
5
in
to
(
7
)
g
iv
es
ℎ
ℎ
=9
2
6
8
0
.
4
8
5
8
.
T
h
en
th
e
v
alu
es
x
an
d
y
f
o
r
2
k
an
d
4
k
r
eso
lu
tio
n
d
is
p
la
y
,
as
p
er
s
tan
d
ar
d
HD
s
ize
is
co
n
s
id
er
ed
in
wo
r
k
is
1
9
2
0
×
1
0
8
0
an
d
3
8
4
0
×
2
1
6
0
r
esp
ec
tiv
el
y
.
T
ab
le
1
s
h
o
ws th
e
ℎ
an
d
h
v
alu
es f
o
r
b
o
th
2
k
an
d
4
k
r
eso
lu
tio
n
.
T
ab
le
1
.
Dis
p
lay
r
eso
lu
tio
n
an
d
co
r
r
esp
o
n
d
in
g
h
y
p
o
ten
u
s
e
R
e
s
o
l
u
t
i
o
n
x
y
ℎ
h
2
K
1
9
2
0
1
0
8
0
2
2
0
2
.
9
0
7
2
0
.
0
2
3
7
7
4
K
3
8
4
0
2
1
6
0
4
4
0
5
.
8
1
5
3
0
.
0
4
7
5
4
T
h
e
r
eso
lu
tio
n
p
ar
am
ete
r
w
in
-
ter
m
s
o
f
n
o
r
m
alize
d
h
y
p
o
te
n
u
s
e
‘
h
’
is
q
u
an
tifie
d
in
(
9
)
.
F
r
o
m
(
9
)
it
is
clea
r
th
at
w
is
to
tally
co
n
tr
o
lled
b
y
ap
p
r
o
p
r
iate
n
o
r
m
alize
d
d
is
tr
ib
u
tio
n
o
f
ℎ
ℎ
an
d
h
v
al
u
es.
=
ℎ
ℎ
ℎ
∈
[
0
,
1
]
(
9
)
3
.
2
.
2
.
Q
ua
ntiz
a
t
io
n
m
a
t
rix
ba
s
ed
o
n po
s
it
io
na
l im
po
rt
a
nce
o
f
t
r
a
ns
f
o
rm
co
e
f
f
icient
T
h
e
b
l
o
ck
wis
e
tr
a
n
s
f
o
r
m
e
d
c
o
ef
f
icien
ts
ar
e
en
er
g
y
co
m
p
ac
ted
p
ix
el
v
alu
es,
wh
er
e
th
e
d
c
an
d
lo
w
f
r
eq
u
e
n
cy
AC
co
m
p
o
n
en
ts
h
a
v
e
to
b
e
r
etain
e
d
with
ca
r
e
a
n
d
h
ig
h
f
r
e
q
u
en
c
y
AC
co
m
p
o
n
en
t
s
h
av
e
to
b
e
s
ca
led
.
T
h
e
p
ix
el
p
o
s
itio
n
b
ased
o
n
t
h
is
r
eq
u
ir
em
e
n
t
ca
n
b
e
ca
lcu
l
ated
b
y
n
o
r
m
alize
d
E
u
clid
ea
n
d
is
tan
ce
p
ar
a
m
eter
(
,
)
g
iv
en
b
y
(
1
0
)
.
(
,
)
=
√
(
1
−
2
)
2
+
(
1
−
2
)
2
(
1
−
)
2
+
(
1
−
)
2
∈
[
0
,
1
]
(
1
0
)
W
h
er
e,
(
i
1
,
j
1
)
=
p
o
s
itio
n
o
f
th
e
DC
co
ef
f
icien
t,
(
i
2
,
j
2
)
=
p
o
s
itio
n
o
f
th
e
c
u
r
r
e
n
t
AC
co
ef
f
icien
t.
(i
max
,
j
max
)
=
p
o
s
itio
n
o
f
th
e
last
AC
co
ef
f
icien
t f
r
o
m
DC
co
ef
f
icien
t f
o
r
th
e
co
n
s
id
er
e
d
b
lo
c
k
s
ize.
Fin
ally
,
a
p
ar
am
eter
M
i,
j
t
h
at
r
elate
s
b
o
t
h
d
is
p
lay
r
eso
lu
tio
n
an
d
p
e
r
ce
p
tu
ally
im
p
o
r
tan
t
tr
an
s
f
o
r
m
co
ef
f
icien
ts
p
o
s
itio
n
in
t
h
e
co
n
s
id
er
ed
b
lo
c
k
s
ize
is
g
iv
en
b
y
(
1
1
)
.
,
(
,
,
)
=
−
[
,
1
+
]
∈
[
0
,
1
]
(
1
1
)
N
o
w
th
is
M
i,
j,
is
ap
p
lied
to
ea
c
h
elem
en
t
o
f
m
atr
ix
H(
x
,
y
)
lo
ca
ted
at
p
o
s
itio
n
(
i,
j)
,
d
e
n
o
ted
as
Hi,
j
d
ef
in
e
d
in
[
21
]
to
p
r
o
d
u
ce
a
n
ad
a
p
tiv
e
2
D
FW
M
H
'
(
i,
j)
.
,
′
=
,
,
(
1
2
)
=
(
,
)
(
1
3
)
Fin
ally
,
th
e
ad
a
p
tiv
e
QM
in
(
1
3
)
ca
n
b
e
o
b
tain
e
d
f
r
o
m
(
1
2
)
an
d
th
e
q
u
an
tizatio
n
p
ar
am
e
ter
wh
ich
co
n
tr
o
ls
q
u
an
tizatio
n
s
tep
s
ize.
T
h
is
m
atr
ix
will
b
e
u
s
ed
to
q
u
an
tize
th
e
tr
an
s
f
o
r
m
c
o
ef
f
ic
ien
ts
b
y
r
ig
h
t
s
h
if
t
o
p
er
atio
n
s
in
s
tead
o
f
r
e
g
u
lar
e
lem
en
t
-
wis
e
d
iv
is
io
n
o
p
e
r
atio
n
s
.
T
h
e
s
am
e
QM
will b
e
s
h
ar
ed
b
etwe
en
f
o
r
war
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
S
eria
l p
a
r
a
llel d
a
ta
flo
w
-
p
i
p
elin
ed
p
r
o
ce
s
s
in
g
a
r
ch
itectu
r
e
b
a
s
ed
a
cc
elera
to
r
… (
S
u
m
a
la
th
a
S
h
iva
r
u
d
r
a
ia
h
)
805
an
d
in
v
er
s
e
p
ath
as
s
h
o
wn
in
F
ig
u
r
e
1
.
Hen
ce
d
u
r
in
g
in
v
er
s
e
q
u
an
tizatio
n
,
ea
ch
t
r
an
s
f
o
r
m
c
o
ef
f
icien
t
lev
el
will
b
e
s
h
if
ted
lef
t b
y
th
e
co
r
r
esp
o
n
d
in
g
v
alu
e
s
p
ec
if
ied
i
n
th
e
QM
in
s
tead
o
f
r
eg
u
lar
m
u
ltip
licatio
n
s
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
th
is
s
ec
tio
n
,
f
ir
s
t
th
e
d
etails
o
f
th
e
e
x
p
er
im
en
tal
s
etu
p
,
lik
e
th
e
to
o
l
u
s
ed
t
o
co
d
e
th
e
p
r
o
p
o
s
ed
d
esig
n
with
in
p
u
t
test
d
ata,
f
o
llo
wed
b
y
f
u
n
ctio
n
al
v
er
i
f
icatio
n
o
f
th
e
2
D
tr
an
s
f
o
r
m
a
n
d
q
u
an
tiz
atio
n
,
ar
e
o
u
tlin
ed
.
On
ce
th
e
f
u
n
ctio
n
al
co
r
r
ec
tn
es
s
o
f
in
d
iv
id
u
al
m
o
d
u
les
an
d
i
n
teg
r
ated
s
y
s
tem
lev
el
is
test
ed
b
y
s
im
u
latio
n
,
t
h
e
d
esig
n
is
s
y
n
th
esized
an
d
im
p
lem
en
ted
o
n
th
e
ta
r
g
eted
So
C
FP
GA
h
ar
d
war
e.
T
h
e
im
p
le
m
en
tatio
n
r
esu
lts
o
n
th
e
h
ar
d
war
e
ar
e
tab
u
lated
in
t
er
m
s
o
f
r
eso
u
r
ce
u
tili
za
tio
n
an
d
o
p
e
r
atin
g
f
r
eq
u
en
cy
.
Als
o
s
h
o
wn
ar
e
co
m
p
a
r
is
o
n
r
esu
lts
o
f
th
e
p
r
o
p
o
s
ed
d
esig
n
with
th
e
r
esear
ch
wo
r
k
s
o
f
o
t
h
er
s
.
4
.
1
.
E
x
perim
ent
a
l set
up
a
nd
f
un
ct
io
na
l v
er
if
ica
t
io
n by
s
i
m
ula
t
io
n
T
h
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
o
f
2
D
tr
an
s
f
o
r
m
atio
n
an
d
q
u
an
tizatio
n
ac
ce
ler
ato
r
m
o
d
u
l
es
an
d
th
e
co
r
r
esp
o
n
d
in
g
test
b
en
c
h
is
co
d
ed
in
Ver
ilo
g
u
s
in
g
Xilin
x
Viv
ad
o
2
0
2
2
.
2
.
At
th
e
s
y
s
tem
lev
el,
all
th
ese
m
o
d
u
les
ar
e
in
s
tan
tiated
in
th
e
to
p
m
o
d
u
le
an
d
co
m
p
iled
t
o
f
it
th
e
tar
g
eted
h
ar
d
war
e.
T
h
e
elab
o
r
ated
co
m
p
lete
ar
ch
itectu
r
e
is
s
h
o
w
n
b
el
o
w
with
two
s
p
lit
s
tag
es
o
f
1
D
-
tr
an
s
f
o
r
m
atio
n
in
Fig
u
r
e
5
a
n
d
2
D
-
tr
an
s
f
o
r
m
atio
n
in
Fig
u
r
e
6
.
Fig
u
r
e
5
.
E
la
b
o
r
ated
s
ch
e
m
atic
o
f
1
D
tr
a
n
s
f
o
r
m
atio
n
s
tag
e
Fig
u
r
e
6
.
E
la
b
o
r
ated
s
ch
e
m
atic
o
f
2
D
tr
a
n
s
f
o
r
m
atio
n
s
tag
e
T
h
e
p
ix
el
v
alu
es
o
f
th
e
im
ag
e
ar
e
r
ea
d
as
a
h
ex
a
d
ec
im
al
ch
a
r
ac
ter
u
s
in
g
f
ile
o
p
e
r
atio
n
s
s
u
p
p
o
r
ted
i
n
Ver
ilo
g
HDL
in
clu
d
ed
in
test
b
en
ch
c
o
d
e.
T
h
e
s
im
u
latio
n
w
av
ef
o
r
m
s
o
b
tain
ed
f
r
o
m
Viv
a
d
o
Simu
lato
r
2
0
2
2
.
2
ar
e
s
h
o
wn
s
tar
tin
g
f
r
o
m
d
ata
r
ea
d
in
to
lin
e
b
u
f
f
er
s
th
en
f
ee
d
i
n
g
to
th
e
tr
an
s
f
o
r
m
atio
n
a
n
d
q
u
an
tizatio
n
m
o
d
u
le
f
o
llo
wed
b
y
wr
itin
g
to
th
e
o
u
t
p
u
t
b
u
f
f
er
.
I
n
th
e
f
o
llo
win
g
s
im
u
latio
n
wav
ef
o
r
m
th
e
s
tag
e
wis
e
o
p
er
atio
n
o
f
th
e
ac
ce
ler
ato
r
m
o
d
u
les
ar
e
illu
s
tr
ated
co
n
s
id
er
in
g
th
e
b
l
o
ck
s
ize
f
o
u
r
an
d
im
ag
e
s
ize
5
1
2
×
5
1
2
with
p
ix
el
s
ize
o
f
eig
h
t
b
it.
T
h
e
lin
e
b
u
f
f
er
s
ar
e
lo
ad
ed
with
in
p
u
t
p
i
x
els,
r
o
w
wis
e
s
eq
u
en
tially
o
n
e
af
ter
th
e
o
th
er
s
h
o
wn
in
Fig
u
r
e
7
(
a)
an
d
r
ea
d
i
n
g
f
o
u
r
-
p
ix
el
v
alu
es
co
lu
m
n
wis
e
f
o
r
f
ee
d
in
g
to
tr
a
n
s
f
o
r
m
atio
n
o
p
er
atio
n
is
s
h
o
wn
in
Fig
u
r
e
7
(
b
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
5
:
798
-
8
0
9
806
(
a)
(
b
)
Fig
u
r
e
7
.
Data
m
o
v
em
en
t
f
r
o
m
s
o
u
r
ce
to
p
r
o
ce
s
s
in
g
u
n
it th
r
o
u
g
h
lin
e
b
u
f
f
er
s
(
a
)
s
er
ial
d
a
ta
wr
itin
g
to
lin
e
b
u
f
f
er
s
an
d
(
b
)
p
a
r
allel
d
ata
r
e
ad
in
g
f
r
o
m
lin
e
b
u
f
f
er
s
As
t
h
e
i
n
p
u
t
im
ag
e
h
as
5
1
2
c
o
l
u
m
n
s
,
to
tal
5
1
2
×
4
=2
0
4
8
p
i
x
e
ls
a
r
e
w
r
itt
e
n
f
i
r
s
t
i
n
t
o
th
e
li
n
e
b
u
f
f
er
s
a
n
d
it
w
ill
b
e
p
r
o
c
ess
e
d
b
y
t
h
e
1
D
t
r
a
n
s
f
o
r
m
m
o
d
u
l
e
as
p
er
e
q
u
a
ti
o
n
(
4
)
.
As
s
h
o
w
n
i
n
F
ig
u
r
e
8
th
e
1
D
t
r
a
n
s
f
o
r
m
ati
o
n
o
f
t
h
e
f
i
r
s
t
4
p
ix
els
ta
k
es
t
h
r
e
e
cl
o
ck
c
y
c
les
,
w
h
e
r
e
m
u
lt
i
p
li
ca
ti
o
n
an
d
ac
cu
m
u
la
ti
o
n
o
f
f
o
u
r
p
a
r
ti
al
p
r
o
d
u
cts
h
a
p
p
e
n
s
in
o
n
e
cl
o
c
k
c
y
c
le
a
n
d
th
e
n
m
o
v
i
n
g
d
ata
t
h
r
o
u
g
h
p
i
p
el
in
e
r
eg
is
te
r
s
ta
k
es
tw
o
cl
o
c
k
c
y
cl
es.
S
o
t
o
g
et
o
n
e
s
e
t
o
f
f
o
u
r
t
r
a
n
s
f
o
r
m
e
d
r
o
w
DC
T
o
u
tp
u
t
,
it
ta
k
es
7
c
lo
c
k
cy
cles
.
N
o
w
e
ac
h
o
f
t
h
ese
1
D
-
DC
T
r
o
w
v
a
lu
es
a
r
e
tr
a
n
s
p
o
s
e
d
u
s
i
n
g
d
e
-
m
u
lt
ip
le
x
er
a
n
d
s
ec
o
n
d
s
ta
g
e
li
n
e
b
u
f
f
e
r
s
as
s
h
o
w
n
i
n
F
i
g
u
r
e
9
.
B
as
e
d
o
n
d
e
-
m
u
lt
ip
le
x
e
r
s
ele
ct
li
n
e
in
p
u
t
,
t
h
e
d
e
-
m
u
l
ti
p
l
ex
er
o
u
t
p
u
t
wil
l
b
e
f
e
d
to
li
n
e
b
u
f
f
er
s
at
e
v
e
r
y
cl
o
ck
cy
cle
,
w
h
i
ch
w
ill
b
e
ac
c
ess
e
d
b
y
2
D
-
tr
an
s
f
o
r
m
ati
o
n
m
o
d
u
le
f
o
r
co
l
u
m
n
-
wis
e
t
r
a
n
s
f
o
r
m
ati
o
n
as
p
er
(
5
)
a
n
d
s
h
o
w
n
i
n
ar
ch
i
tect
u
r
e
Fi
g
u
r
e
2
.
A
t
ev
er
y
cl
o
c
k
c
y
c
le
t
h
e
1
D
t
r
a
n
s
f
o
r
m
e
d
p
i
x
el
g
ets
m
u
lti
p
l
ie
d
w
ith
t
r
a
n
s
p
o
s
e
d
k
e
r
n
el
c
o
ef
f
i
cie
n
ts
t
o
p
r
o
d
u
c
e
f
o
u
r
p
a
r
ti
al
p
r
o
d
u
ct
co
lu
m
n
s
,
s
h
o
w
n
i
n
Fi
g
u
r
e
1
0
.
A
ll
th
ese
p
ar
tia
l
p
r
o
d
u
c
ts
ar
e
a
cc
u
m
u
lat
ed
r
o
w
wis
e
at
f
o
u
r
t
h
cl
o
c
k
cy
cles
t
o
o
u
t
p
u
t
f
o
u
r
2
D
-
t
r
an
s
f
o
r
m
e
d
r
esu
lts
.
Fig
u
r
e
8
.
1
D
tr
an
s
f
o
r
m
atio
n
o
u
tp
u
t
Fig
u
r
e
9
.
1
D
tr
an
s
f
o
r
m
atio
n
o
u
tp
u
t tr
an
s
p
o
s
e
u
s
in
g
DE
MU
X
an
d
s
ec
o
n
d
s
tag
e
lin
e
-
b
u
f
f
er
s
Fig
u
r
e
1
0
.
Ser
ial
an
d
p
ar
allel
p
r
o
ce
s
s
in
g
o
f
2
D
-
tr
an
s
f
o
r
m
ati
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
S
eria
l p
a
r
a
llel d
a
ta
flo
w
-
p
i
p
elin
ed
p
r
o
ce
s
s
in
g
a
r
ch
itectu
r
e
b
a
s
ed
a
cc
elera
to
r
… (
S
u
m
a
la
th
a
S
h
iva
r
u
d
r
a
ia
h
)
807
Hen
ce
in
t
h
e
p
r
o
p
o
s
ed
ar
c
h
itectu
r
e,
b
ased
o
n
th
e
s
elec
ted
b
lo
ck
s
ize,
th
e
m
u
ltip
le
p
ix
el
d
ata,
m
ad
e
av
ailab
le
n
ea
r
th
e
p
r
o
ce
s
s
in
g
m
o
d
u
le
with
in
lin
e
b
u
f
f
er
s
an
d
th
en
th
is
d
ata
will
b
e
p
r
o
ce
s
s
ed
to
h
av
e
m
u
ltip
le
o
u
tp
u
ts
in
p
ar
allel.
T
h
e
QM
o
b
tain
ed
f
r
o
m
(
1
3
)
f
o
r
th
e
s
elec
ted
q
u
an
tizatio
n
p
a
r
am
eter
will
b
e
u
s
ed
to
q
u
an
tize
ev
er
y
2
D
-
t
r
an
s
f
o
r
m
ed
o
u
tp
u
t
p
ix
el
b
y
r
ig
h
t
s
h
if
t
o
p
e
r
atio
n
.
Hen
ce
th
e
p
r
o
p
o
s
ed
d
esig
n
h
a
s
f
o
u
r
p
ip
elin
e
s
tag
es
lik
e
1
D
tr
a
n
s
f
o
r
m
atio
n
,
d
ata
tr
an
s
p
o
s
e,
2
D
tr
an
s
f
o
r
m
ati
o
n
an
d
q
u
an
tizatio
n
.
I
n
th
e
p
r
o
p
o
s
e
d
p
ip
elin
e
ar
ch
itectu
r
e,
f
o
r
th
e
s
elec
ted
4
×
4
b
lo
c
k
s
ize
p
r
o
ce
s
s
in
g
s
h
o
wn
in
Fig
u
r
e
1
1
h
as
a
laten
c
y
o
f
1
5
clo
ck
c
y
cles
f
r
o
m
i
n
p
u
t
t
o
o
u
tp
u
t.
T
h
e
T
ab
le
2
s
h
o
ws
th
e
to
tal
clo
ck
cy
cles
r
eq
u
ir
e
d
to
p
r
o
ce
s
s
tr
an
s
f
o
r
m
atio
n
an
d
q
u
an
tizatio
n
o
f
d
if
f
er
en
t
b
lo
ck
s
izes f
r
o
m
4
×
4
to
3
2
×
3
2
with
p
ip
elin
e
laten
cy
o
f
ea
c
h
s
tag
e.
Fig
u
r
e
1
1
.
2
D
-
t
r
an
s
f
o
r
m
atio
n
an
d
q
u
an
tizatio
n
p
i
p
elin
ed
o
p
er
atio
n
o
u
t
p
u
t f
o
r
b
lo
c
k
s
ize
4
×
4
T
ab
le
2
.
Nu
m
b
er
o
f
clo
ck
cy
cl
es r
eq
u
ir
ed
f
o
r
2
D
-
DC
T
La
t
e
n
c
y
o
f
p
i
p
e
l
i
n
e
s
t
a
g
e
s
B
l
o
c
k
s
i
z
e
4
×
4
B
l
o
c
k
s
i
z
e
8
×
8
B
l
o
c
k
s
i
z
e
1
6
×
16
B
l
o
c
k
s
i
z
e
3
2
×
32
D
C
T
-
1
I
/
p
t
o
D
C
T
-
1
o
/
p
3
3
3
3
Tr
a
n
s
p
o
se
1
D
-
D
C
T
o
/
p
4
4
4
4
D
C
T
-
2
i
/
p
t
o
D
C
T
-
2
o
/
p
4
+
1
=
5
8
+
1
=
9
1
6
+
1
=
1
7
3
2
+
1
=
3
3
Q
u
a
n
t
i
z
a
t
i
o
n
a
n
d
f
i
n
a
l
o
/
p
t
h
r
o
u
g
h
o
u
t
p
u
t
b
u
f
f
e
r
3
3
3
3
To
t
a
l
c
l
o
c
k
c
y
c
l
e
s
15
19
27
43
4.
2
.
Sy
nthesis
re
s
ults a
nd
d
i
s
cus
s
io
n
T
h
e
Xilin
x
Viv
ad
o
2
0
2
2
.
2
s
y
n
th
esis
to
o
l
is
u
s
ed
to
s
y
n
th
esize
th
e
p
r
o
p
o
s
ed
s
ca
lab
le
tr
an
s
f
o
r
m
atio
n
an
d
q
u
an
tizatio
n
p
i
p
elin
e
ar
ch
itectu
r
e
wh
ich
d
o
es
th
e
u
n
if
ie
d
f
o
r
war
d
an
d
in
v
e
r
s
e
o
p
er
ati
o
n
s
o
f
in
p
u
t
d
ata
i
n
ter
m
s
o
f
b
lo
c
k
s
izes
4
×
4
to
3
2
×
3
2
.
T
h
e
s
y
n
th
esis
an
d
im
p
le
m
en
tatio
n
o
f
th
e
p
r
o
p
o
s
ed
a
r
c
h
itectu
r
e
is
tar
g
eted
to
two
d
if
f
e
r
en
t
So
C
FP
GA
d
ev
ices
Z
y
n
q
Z
C
7
0
2
an
d
Z
y
n
q
Ultr
ascale+
Z
C
U
-
104
r
esp
ec
t
iv
ely
.
T
h
e
T
ab
le
3
s
h
o
ws
th
at
th
e
p
r
o
p
o
s
ed
p
i
p
elin
e
ar
ch
itectu
r
e
h
as
a
tr
ad
e
-
o
f
f
b
etwe
en
p
o
wer
co
n
s
u
m
p
tio
n
a
n
d
s
p
ee
d
,
wh
er
e
o
n
th
e
lo
w
e
n
d
FP
GA
Z
C
7
0
2
c
o
n
s
u
m
es
less
en
er
g
y
th
an
h
ig
h
e
n
d
Z
C
U
-
1
0
4
FP
GA
o
f
av
er
ag
e
p
o
wer
c
o
n
s
u
m
p
tio
n
0
.
1
5
W.
T
h
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
ar
ch
itectu
r
e
o
n
h
i
g
h
e
n
d
FP
GA
is
alm
o
s
t
twice
th
at
o
f
l
o
w
en
d
FP
GA.
Als
o
n
o
ticed
th
at
th
e
o
p
er
atin
g
f
r
e
q
u
en
cy
o
f
th
e
a
r
ch
itectu
r
e
d
ec
r
ea
s
es a
s
th
e
b
lo
ck
s
ize
in
cr
ea
s
es.
T
h
e
p
r
o
p
o
s
ed
p
ip
elin
ed
a
r
ch
i
tectu
r
e
is
co
m
p
ar
e
d
with
o
th
er
h
ar
d
war
e
im
p
lem
en
tatio
n
s
s
h
o
wn
in
T
ab
le
4
.
T
h
e
r
eso
u
r
c
e
u
tili
za
tio
n
an
d
n
u
m
b
er
o
f
clo
ck
cy
cle
s
r
eq
u
ir
ed
to
p
r
o
ce
s
s
th
e
m
ax
im
u
m
b
lo
ck
s
ize
o
f
32
×
3
2
in
t
h
e
p
r
o
p
o
s
ed
ar
c
h
itectu
r
e
is
co
m
p
a
r
ed
with
o
th
er
s
wo
r
k
.
T
h
e
h
ar
d
wa
r
e
im
p
lem
en
tatio
n
o
f
[
5
]
co
m
p
u
tes
2
D
-
DC
T
o
p
er
atio
n
s
b
y
ev
e
n
-
o
d
d
d
ec
o
m
p
o
s
itio
n
b
ased
b
u
tter
f
ly
s
tr
u
ctu
r
e
an
d
u
s
es
alm
o
s
t
8
-
tim
es
m
o
r
e
DSP
elem
en
ts
th
an
o
u
r
im
p
lem
en
tatio
n
.
Als
o
it
tak
es
[
5
]
,
5
0
0
clo
ck
cy
cles
to
p
r
o
ce
s
s
3
2
×
3
2
b
lo
ck
s
ize
DC
T
,
wh
er
e
o
u
r
ar
ch
itectu
r
e
r
eq
u
ir
es
o
n
ly
4
3
clo
ck
cy
cles.
T
h
e
r
e
d
u
ce
d
n
u
m
b
er
o
f
clo
c
k
cy
cles
in
th
e
p
r
o
p
o
s
e
d
ar
ch
itectu
r
e
is
b
ec
au
s
e
o
f
two
im
p
o
r
tan
t
tech
n
iq
u
es,
f
ir
s
t
o
n
e
th
e
r
eq
u
ir
ed
am
o
u
n
t
o
f
d
at
a
f
o
r
p
r
o
ce
s
s
in
g
is
Evaluation Warning : The document was created with Spire.PDF for Python.