Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
Vol
.
5
,
No
. 3,
J
une
2
0
1
5
,
pp
. 56
2~
56
8
I
S
SN
: 208
8-8
7
0
8
5
62
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Implementation of Cloth Simula
tion Using Parallel Computing
on M
o
bile Device
Jae
H
o
n
g
Jeon*,
Se Don
g
Min**,
and Min
Hong***
* Department of
Computer Scien
ce, Gradua
te Sch
ool, Soonchunh
yang University
,
Korea
** Departmen
t
o
f
Medical IT En
gineer
ing, Soonchunh
y
a
ng
University
, Korea
*** Departmen
t
of Computer Sof
t
ware
E
ngin
eering, Soonchunh
yang University
,
Korea
Article Info
A
B
STRAC
T
Article histo
r
y:
Received
Ja
n 13, 2015
Rev
i
sed
Ap
r
20
, 20
15
Accepte
d
May 5, 2015
Ph
y
s
ically
based modeli
ng and simulation is
an important technique for
deform
able obj
e
c
t sim
u
lat
i
on, w
h
ich is wide
l
y
u
s
ed to repr
esent
the re
alist
i
c
shape ch
ange
an
d movement of
objects
for m
obile gam
e
o
r
3D sim
u
lation
.
However, th
ey
require the hig
h
com
putational cost for
representing th
e
ph
y
s
ical pheno
menon on deformable obj
ects when it applied
on mobile
device. In th
is p
a
per, we d
e
sign
ed
and
im
plem
e
n
ted
the
clo
t
h si
m
u
lation for
deformable object simulation using the
parallel
techn
i
que on mobile device
to optim
iz
e
the
com
putation
a
l b
u
rden. W
e
espe
cia
l
l
y
appli
e
d G
P
U parall
e
l
techn
i
que for th
e integr
ation solving
process such as Euler
,
Midpoint, 4th-
order Runge-K
utta method
to
estimate
the particles
'
n
e
xt status
using
positions and velocit
i
es. Also we appli
e
d m
u
lti-t
h
read par
a
ll
el te
chnique fo
r
cal
cula
ting
the
s
p
ring force
.
T
h
en we
com
p
ar
ed th
e perfo
rm
ance of
e
ach
integr
ation meth
ods between und
er onl
y
CPU and
CPU with GPU on m
obile
devic
e
.
Als
o
we
com
p
ared th
e
co
m
pu
ting time of
spring calculatio
n between
onl
y CPU and us
ing CPU m
u
lti-t
h
read.
Keyword:
C
l
ot
h si
m
u
l
a
t
i
on
GP
GP
U
Mass spring sy
ste
m
Sha
d
er
Tran
sf
orm
feedbac
k
Copyright ©
201
4 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Min
Ho
ng,
Depa
rt
m
e
nt
of
C
o
m
put
er S
o
ft
ware
En
gi
nee
r
i
n
g
,
Soo
n
c
hu
nh
yang
U
n
i
v
er
sity,
U
n
it
5
0
7
,
Mu
lt
i
m
ed
ia Bld
., So
on
ch
unh
yang
U
n
i
v
., Eu
m
n
ae-
r
i
,
Sin
c
h
a
ng-myeo
n
,
A
s
an
-
s
i,
C
h
u
n
g
che
o
ng
n
a
m
-
do
, 33
6-
7
4
5
R
e
p. of
K
o
re
a
Em
ail: m
hong@sch.ac.kr
1.
INTRODUCTION
Al
t
h
o
u
gh
va
ri
ous m
obi
l
e
ap
pl
i
cat
i
ons
have
been i
m
pl
em
ent
e
d
wi
t
h
rece
nt
ad
vance
d
m
obi
l
e
de
vi
c
e
t
echn
o
l
o
gi
es,
m
o
st
appl
i
cat
ions a
r
e base
d
on
2D c
ont
e
n
t
s
due t
o
t
h
e l
i
m
i
t
e
d perf
orm
a
nce o
f
m
obi
l
e
devi
ces
.
Thu
s
, t
h
e im
p
l
e
m
en
tatio
n
of
g
a
m
e
, an
i
m
ati
o
n,
v
i
rtu
a
l
reality, au
g
m
en
ted
reality, an
d ad
v
e
rtisem
en
t
wh
i
c
h
requ
ire
realistic represen
tatio
n
of
3
D
ob
jects is d
i
ffi
cu
lt to
im
p
l
e
m
en
t o
n
m
o
b
ile d
e
v
i
ces. In
ad
d
ition
,
the
adve
nt
o
f
rece
nt
HM
D
(Hea
d
M
o
u
n
t
e
d
Di
sp
l
a
y
)
devi
ces s
u
ch as Oc
ul
us l
i
fe an
d Gea
r
V
R
have
been
re
cei
ved
th
e p
u
b
lic att
e
n
tio
n
in
VR
(Virtu
al Reality) wo
rld
.
Th
e real ob
j
ect
s th
at h
a
v
e
rig
i
d
or d
e
formab
le
ch
aracteristics fo
r realistic represen
tatio
n o
f
m
o
v
e
m
e
nt
and i
n
t
e
ract
i
on
bet
w
ee
n o
b
ject
s
hi
g
h
l
y
req
u
i
r
e
p
h
y
sically-b
ased
sim
u
latio
n
.
A 3D de
formable object sim
u
la
tion can
represe
n
t the deform
ed
obje
cts like real objects
,
but it
req
u
i
r
es t
h
e
hi
gh
com
put
at
i
o
nal
p
o
w
er
due
to re
pres
ent t
h
e 3D
object
worl
d
or related
calcu
latio
n
.
It h
a
s
to
co
m
p
u
t
e lo
ts o
f
ph
ysically
related
calcu
latio
n
s
in
eac
h sim
u
lation time step, so t
h
e physically base
d
si
m
u
latio
n
is
no
t triv
ial t
o
p
e
rfo
r
m
o
n
m
o
b
ile d
e
v
i
ces.
To
s
o
lv
e th
e
s
e pr
ob
le
ms
,
ma
n
y
r
e
s
e
a
r
c
h
es
h
a
v
e
b
een
wid
e
ly stu
d
i
ed o
n
n
u
m
erical
in
teg
r
ation
m
e
th
od
s, co
llis
io
n
ch
eck
and
resp
on
se m
e
th
o
d
s [1
], m
u
lti-co
re and
paral
l
e
l
ap
pr
oa
ches [
2
]
o
n
m
obi
l
e
de
vi
ces.
Ho
we
ver m
o
st
researc
h
es
ha
ve bee
n
f
o
c
u
s
e
d o
n
t
h
e
reg
u
l
ar PC
envi
ro
nm
ent
s
not
m
obi
l
e
e
n
v
i
ro
nm
ent
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
5, No
. 3,
J
u
ne 2
0
1
5
:
56
2 – 5
6
8
56
3
Recent m
obile de
vices
provi
de the
GPU that has
re
latively low cl
ock speed,
bu
t it provides
lots
of
ALUs (Arith
metic Lo
g
i
c Un
i
t
) wh
ich
can
be ap
p
lied
to
parallel p
r
o
cessi
n
g
. Ho
wev
e
r, th
ere are no
official
p
a
rallel p
r
o
c
essin
g
related
lib
raries
for mo
b
ile d
e
v
i
ces
u
n
til no
w
wh
i
c
h
can
con
v
e
n
i
en
tly su
ppo
rt th
e
i
m
p
l
e
m
en
tatio
n
of m
o
b
ile app
licatio
n
.
Th
e
GPU is si
m
u
lt
an
eou
s
ly workin
g
with
ALUs, so wh
en
GPU go
t
wo
rk
pr
ocess
,
i
t
separat
e
s an
d l
a
unc
hes t
h
e
num
ber o
f
AL
Us
an
d pe
rf
or
m
s
t
oget
h
er i
n
sam
e
t
i
m
e
. This i
s
we
call GPGPU (Gene
r
al-Purpose com
puting
on
Gra
p
hics Processing Units).
In t
h
i
s
pa
pe
r,
we p
r
o
p
o
sed a
nd i
m
pl
em
ent
e
d t
h
e p
h
y
s
i
cal
l
y
based cl
ot
h s
i
m
u
l
a
t
i
on [3]
u
s
i
ng a ve
rt
ex
sh
ad
er that com
p
u
t
es lo
ts o
f
p
a
rticle po
sitio
n
s
w
ith
G
P
GPU
sim
u
ltan
e
ou
sly.
W
e
pr
opo
sed
a n
e
w
al
g
o
r
ith
m
that using a transform
feed
back [4] which
can capt
u
re t
h
e data from
a sha
d
er a
nd ca
n re
use them
in ne
xt
si
m
u
latio
n
ti
me. Also
we
prop
o
s
ed
n
e
w cloth
sim
u
lat
i
o
n
d
a
ta stru
ctu
r
e
an
d
m
u
lti-th
read
[5
] [6
] m
e
t
h
od
for
calcu
latin
g
t
h
e spring
fo
rces. In
ad
d
ition
,
we
p
e
rform
e
d
th
e clo
t
h
si
m
u
la
tio
n
using
o
n
l
y CPU
an
d th
e
pr
o
pose
d
GP
U
paral
l
e
l
com
put
i
ng a
n
d com
p
are
d
t
h
e
per
f
o
rm
ance of
sp
ri
n
g
cal
cul
a
t
i
o
n t
i
m
e
usi
ng
m
u
lt
i
-
t
h
rea
d
C
P
U a
n
d
onl
y
C
P
U.
2.
CLOTH SIMULATION USIN
G PARAL
LEL COMPUTING
The c
o
m
put
at
ion
o
f
p
r
o
p
o
se
d cl
ot
h
si
m
u
l
a
t
i
on u
s
i
n
g G
P
GP
U m
e
t
hod
can be
cl
assi
fi
ed i
n
t
o
t
w
o
part
s:
C
P
U c
o
m
put
at
i
on an
d
GP
U c
o
m
put
a
t
i
on.
The
ba
si
c cl
ot
h
si
m
u
l
a
t
i
on
p
r
ope
rt
i
e
s s
u
ch
as
n
ode
s,
spri
ng
co
nn
ection
s
, ex
tern
al forces
an
d
so
on
are i
n
itialized
a
n
d
sto
r
ed
on
CPU. Th
en
it co
m
p
u
t
es all sp
ring
forces
and s
u
m
up these force
s
with external forc
es at each
nodes using CPU. In ne
xt
step, it co
m
putes the next
status of
node
positions using
GP
GPU using
GPU. Al
l ne
xt positions of each
node are inde
pe
nde
ntly
com
put
ed by
GP
GP
U,
s
o
i
t
has
t
o
be
ar
ra
n
g
ed
with
s
u
itab
l
e fo
rm
at for
G
P
U a
n
d CP
U.
2.
1 Da
ta
S
t
ruc
t
ure of
Cl
ot
h Si
mul
a
ti
on
Data stru
cture of clo
t
h
sim
u
latio
n
in ou
r
p
a
p
e
r
is
d
i
fferen
t fro
m
th
e t
r
ad
ition
a
l appro
ach.
Nod
e
in
fo
rm
atio
n
in
trad
itio
n
a
l clo
t
h
sim
u
latio
n
in
clu
d
e
s t
h
e po
si
tio
n
,
v
e
lo
city, an
d
fo
rce in
each
un
it. Th
is syste
m
sh
ou
l
d
co
n
t
i
n
uo
u
s
ly m
a
n
a
g
e
th
e d
a
ta tran
smissio
n
to
send
th
ese d
a
ta to GPU m
e
m
o
ry
in
ev
ery ti
m
e
step
. In
th
e p
r
op
o
s
ed
alg
o
rith
m
,
th
e p
o
s
itio
n, v
e
lo
cit
y
, an
d
fo
rce inform
at
io
n
are
man
a
g
e
d
with
ex
actly sa
m
e
f
o
rm
at
i
n
G
P
U
m
e
m
o
ry
by
l
i
s
t
an
d
whe
n
t
h
ese i
n
f
o
rm
at
i
on are
r
e
qui
red
t
o
be c
a
l
c
ul
at
ed, t
h
ey
are
refe
rre
d
us
i
ng l
i
s
t
by
p
o
i
n
t
e
r. T
h
ere
f
o
r
e,
t
h
e
pr
o
pose
d
dat
a
st
ruct
ur
e
can pre
v
e
n
t the
unnecessa
ry
data tran
sm
issi
o
n
for
rearrang
em
en
t o
f
po
sitio
n,
v
e
lo
city, fo
rce and
m
a
ss in
ev
ery ti
m
e
step
. Fig
u
re
1
sho
w
s th
e co
m
p
ariso
n
o
f
d
a
ta
stru
cture
b
e
tween
th
e trad
itio
nal clo
t
h
sim
u
latio
n
an
d th
e pro
p
o
s
ed
cl
o
t
h
si
m
u
la
tio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Imp
l
emen
ta
tion
o
f
Clo
t
h S
i
mu
la
tion
Using
Pa
ra
llel Co
mpu
tin
g on
Mob
ile Device
(Min
H
ong
)
56
4
Fi
gu
re
1.
Dat
a
st
ruct
u
r
e c
o
m
p
ari
s
o
n
bet
w
ee
n
t
h
e t
r
a
d
i
t
i
onal
and
p
r
op
ose
d
c
l
ot
h si
m
u
l
a
t
i
on
2.
2 Cal
c
ul
ate
of
Spri
n
g
F
o
r
ce
usi
n
g Mul
t
i
-
T
h
rea
d of
C
P
U
Gene
ral
l
y
, cal
cul
a
t
i
on
of s
p
r
i
ng f
o
rce
s
i
n
cl
ot
h si
m
u
l
a
ti
on
req
u
i
r
es l
o
t
s
of c
o
m
put
at
i
onal
t
i
m
e
and
th
e trad
itio
n
a
l
clo
t
h
sim
u
latio
n
u
s
ed on
ly one CPU
for
t
h
is com
putation t
h
at leads t
h
e
decreased sim
u
lation
p
e
rform
a
n
ce. To
so
l
v
e th
e p
r
ob
lem
,
th
e
p
r
op
o
s
ed
m
e
th
o
d
u
tilized
th
e
m
u
lti-th
read
ap
pro
ach
on
CPU to
calcu
late th
e sp
ri
n
g
forces in p
a
rallel
m
a
n
n
er. Th
e propo
sed
algo
rith
m
sp
lits n
spring
s
in
to
16
thread
g
r
ou
p
s
and eac
h gr
o
u
p
i
nde
pe
nde
nt
l
y
cal
cul
a
t
e
s spri
n
g
fo
rces wi
t
h
i
t
s
own t
h
re
ad.
Whe
n
t
h
e cal
cul
a
t
i
on o
f
spri
ng
forces is fi
n
i
shed
in
ev
ery t
h
read
, t
h
e proposed
algorith
m
calcu
lates th
e
n
e
x
t
status of
n
o
d
e
p
o
sitio
n
s
u
s
ing
num
erical integration. Si
nce each spring
is connected
with two
nodes, the calculated spring forces are
store
d
in ass
o
ciated
node i
n
form
ation. Beca
use s
e
veral t
h
rea
d
s
can acces
s to sam
e
node at
the sam
e
time, we
appl
i
e
d
t
h
e m
u
t
e
x (m
ut
ual
ex
cl
usi
o
n) t
o
pre
v
ent
t
h
i
s
p
r
obl
em
. Fi
gu
re
2 s
h
o
w
s t
h
e
fl
o
w
chart
of
i
m
pl
em
ent
e
d
clo
t
h
sim
u
latio
n
with
m
u
lti-th
read of CPU an
d with on
e C
P
U.
Po
s
i
t
i
o
n
(Floa
t
*
[3])
Ve
l
o
c
i
t
y
(Floa
t
*
[3])
Fo
rc
e
(Floa
t
*
[3])
Mass
(Floa
t
*
)
~
~
~
~
D
at
Po
s
i
t
i
o
n
(Floa
t
*
[3])
Ve
l
o
c
i
t
y
(Floa
t
*
[3])
Fo
rc
e
(Floa
t
*
[3])
Mass
(Floa
t
*
)
~
~
~
~
D
at
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
5, No
. 3,
J
u
ne 2
0
1
5
:
56
2 – 5
6
8
56
5
Fig
u
re
2
.
Flowch
art
o
f
clo
t
h
si
m
u
latio
n
u
s
i
n
g
m
u
lti-th
read
CPU an
d
o
n
e
CPU
2.
3
Cl
o
t
h Si
m
u
l
a
ti
on
usi
n
g
t
h
e T
r
a
n
sf
orm
Feedb
a
ck
B
u
ffer
The C
P
U pa
ss
es al
l
no
de dat
a
i
n
t
o
G
P
U m
e
m
o
ry
, and
vert
ex s
h
ade
r
c
onc
ur
rent
l
y
cal
cul
a
t
e
s t
h
e ne
xt
n
o
d
e
po
sition
s
u
s
ing
GPGPU an
d
th
en
upd
ated
n
o
d
e
p
o
s
iti
o
n
d
a
ta sho
u
l
d b
e
tran
sferred to
in
to
CPU.
In
th
is
r
e
s
e
a
r
ch
, w
e
ap
p
lie
d th
e tr
ans
f
or
m f
e
e
d
b
a
ck
b
u
f
f
e
r
th
at i
s
rece
ntly suppos
ed
by
o
p
en
GL
ES
.
Usin
g th
e
trans
f
orm
feedback
buffer,
we can
cap
ture th
e no
d
e
d
a
ta after calcu
latio
n o
n
GPU and
th
en
p
u
sh
th
ese d
a
ta
in
to
th
e tran
sform feed
b
a
ck
buffer. Therefore, the im
pl
em
ent
e
d cl
ot
h si
m
u
l
a
t
i
on wi
t
h
t
h
e t
r
ansf
orm
feedba
c
k
buffe
r ca
n efficiently transfer the
related data fr
om
GPU t
o
C
P
U
.
Fi
gu
re
3 sh
o
w
s t
h
e
fl
o
w
c
h
art
of
im
pl
em
ent
e
d cl
ot
h si
m
u
l
a
t
i
on usi
n
g
GP
U.
Fi
gu
re
3.
Fl
o
w
chart
of
t
h
e i
m
pl
em
ent
e
d cl
ot
h si
m
u
l
a
t
i
on w
i
t
h
GP
U
3.
R
E
SU
LTS AN
D ANA
LY
SIS
Fi
gu
re
4 s
h
o
w
s t
h
e
res
u
l
t
of e
x
peri
m
e
nt
al
t
e
st
whi
c
h
was
per
f
o
r
m
e
d
on
IPa
d
Ai
r
R
e
t
i
n
a.
We
per
f
o
r
m
e
d Sem
i
-Eul
er, M
i
d
poi
nt
, 4t
h
-
or
de
r R
u
n
g
e
-
K
u
t
t
a
m
e
t
hod t
o
co
m
p
are t
h
e perf
orm
a
nce of
G
P
U an
d
CPU
en
v
i
r
o
n
m
en
ts
u
n
d
e
r fr
om
3
,
6
0
0
to 250,00
0 v
e
r
tices.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Imp
l
emen
ta
tion
o
f
Clo
t
h S
i
mu
la
tion
Using
Pa
ra
llel Co
mpu
tin
g on
Mob
ile Device
(Min
H
ong
)
56
6
Fi
gu
re
4.
S
n
a
p
shot
s
o
f
i
m
pl
em
ent
e
d cl
ot
h si
m
u
l
a
t
i
o
n
As a resu
lt, t
h
e p
r
op
o
s
ed
GPU p
a
rallel techn
i
qu
e wit
h
the
trans
f
orm
feedback
buffer is
m
u
ch faster
t
h
an
onl
y
C
P
U
,
whe
n
t
h
e
nu
m
b
er of n
odes
i
s
over t
h
e s
p
eci
fi
c brea
ki
n
g
poi
nt
. Si
nce c
l
ot
h si
m
u
l
a
t
i
on wi
t
h
GPU
p
a
rallel co
m
p
u
tin
g
h
a
s to
tran
sfer t
h
e
d
a
ta fro
m
CPU to GPU m
e
m
o
ry, it req
u
i
res th
e tran
sfer
laten
c
y.
Ho
we
ver
,
whe
n
t
h
e
num
ber
of
n
o
d
es i
s
o
v
e
r
6,
00
0
ve
rt
i
ces, t
h
e
pr
op
os
ed m
e
t
hod
i
s
m
u
ch fast
er
t
h
an
o
n
l
y
CPU m
e
thod. In a
d
dition, 4th-orde
r R
u
nge
-
Kutta
m
e
thod which
requires m
o
re
com
p
lex com
puting
ope
rat
i
o
ns
has
m
o
re adva
nt
ag
e wi
t
h
GP
U.
F
i
gu
re 5
sh
o
w
s
t
h
e pe
rf
orm
a
nc
e res
u
l
t
(m
s) o
f
com
p
ari
s
o
n
wi
t
h
3
di
ffe
re
nt
i
n
t
e
g
r
at
i
on m
e
t
hods
usi
n
g
GP
U a
n
d
C
P
U.
Fi
gu
re
5.
The
per
f
o
r
m
a
nce re
sul
t
wi
t
h
di
f
f
er
ent
i
n
t
e
g
r
at
i
o
n
m
e
t
hods
wi
t
h
C
P
U a
n
d
G
P
U
Th
e propo
sed
m
u
l
ti-th
read
C
P
U p
a
rallel tec
h
n
i
q
u
e
m
e
th
o
d
is
m
u
ch
faster th
an
CPU on
l
y
, wh
en
th
e
n
u
m
b
e
r of spri
n
g
s
is ov
er 40
,0
00
. Spring
force calcu
latio
ns can
b
e
qu
ickly p
r
o
cessed
by sp
littin
g
th
em in
to
m
u
l
ti-th
read
C
P
U. Alth
oug
h
th
ere
is
no si
g
n
i
f
i
cant
i
m
pro
v
em
ent
of per
f
o
rm
ance whe
n
t
h
e num
ber of
spri
n
g
i
s
l
o
w, w
h
en t
h
e n
u
m
b
er of
spri
ngs i
s
get
t
i
ng i
n
c
r
ease
d
, t
h
e m
u
l
t
i
-
t
h
rea
d
di
vi
des t
h
e s
p
ri
ngs i
n
t
o
1
6
gr
o
ups
an
d
q
u
i
ck
ly calcu
lates th
em u
s
i
n
g m
u
lti-c
o
re in m
o
b
ile
d
e
v
i
ce.
Th
erefore, we b
e
liev
e
t
h
at th
e pro
p
o
s
ed
m
e
t
hod i
s
wel
l
sui
t
a
bl
e
fo
r s
o
m
e
sim
u
l
a
t
i
on appl
i
cat
i
o
ns t
h
at
req
u
i
r
e
pl
au
si
bl
e real
-t
i
m
e
per
f
o
r
m
a
nce o
f
cl
o
t
h
si
m
u
latio
n
fo
r
m
o
b
ile d
e
v
i
ces. Figu
re 6 shows th
e resu
lt of p
e
rform
a
n
ce test fo
r clo
t
h
si
m
u
la
tio
n
wh
en m
u
lt
i-
th
read
CPU
was app
lied
t
o
th
e calcu
latio
n of
sp
ri
n
g
fo
rces.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
5, No
. 3,
J
u
ne 2
0
1
5
:
56
2 – 5
6
8
56
7
Fig
u
re
6
.
Th
e
p
e
rform
a
n
ce resu
lt of sp
ri
n
g
fo
rc
e calcu
lation
u
s
ing
m
u
lti-th
read
CPU and on
ly CPU
4.
CO
NCL
USI
O
NS
In t
h
i
s
pa
pe
r,
we p
r
op
ose
d
t
h
e cl
ot
h si
m
u
lat
i
on
usi
n
g
par
a
l
l
e
l
co
m
put
i
n
g m
e
t
hods t
h
at
are b
a
sed
o
n
th
e m
u
lti-th
read
with
C
P
U and
th
e tr
an
sform feedb
a
ck
bu
ffer
with
GPU. Using
th
e
tr
an
sform
feed
b
a
ck
b
u
ffer
whi
c
h can ca
pt
ures
no
de i
n
fo
rm
ati
on, w
h
e
n
shade
r
i
s
wo
r
k
i
n
g o
n
m
obi
l
e
devi
ce.
T
h
e m
u
lti-thread C
P
U can
d
i
v
i
d
e
and
ex
ecu
t
e th
e calcu
latio
n
o
f
spring forces
with
some o
f
g
r
o
u
p
s
.
To
u
p
d
a
te th
e
p
o
s
ition
an
d velo
city
of no
des, we use
d
Sem
i
-Eul
er,
M
i
d
poi
nt
,
4t
h
-
o
r
de
r Rung
e-Ku
tta in
tegratio
n
m
e
th
o
d
s b
a
sed
on
CPU and
GP
U. T
h
e p
r
o
pos
ed
GP
U pa
r
a
l
l
e
l
t
echni
que
i
s
m
u
ch fast
er
th
an
CPU,
wh
en
th
e nu
m
b
er of no
d
e
s is relativ
ely
hi
g
h
en
o
u
g
h
.
Al
so t
h
e m
u
l
t
i
-t
hrea
d C
P
U p
a
ral
l
e
l
t
echni
q
u
e re
duce
s
t
h
e
spri
ng
fo
rce co
m
put
i
ng t
i
m
e
in cl
ot
h
sim
u
l
a
t
i
on. I
n
t
h
i
s
pa
per, al
t
h
ou
g
h
we c
o
nfi
r
m
e
d t
h
e per
f
o
r
m
a
nce of t
h
e
C
P
U an
d
GP
U
paral
l
e
l
com
put
i
n
g
app
r
oach
, we di
d n
o
t
com
b
i
n
e t
w
o
pr
op
os
ed m
e
t
hod
s y
e
t
.
The i
n
t
e
grat
ed cl
ot
h si
m
u
lat
i
on usi
ng b
o
t
h
C
P
U
and GPU pa
ral
l
el processing
is exp
ected t
o
provide m
u
ch
im
proved pe
rf
orm
a
nce o
f
cl
ot
h
si
m
u
l
a
ti
on,
w
h
e
n
th
e spring
force co
m
p
u
t
atio
n
wit
h
m
u
lti-t
h
read
CPU and
wit
h
GPU
usin
g
t
r
an
sfo
r
m feedb
a
ck
b
u
ffer are
co
m
b
in
ed
. In
ad
d
ition
,
real-ti
m
e
d
e
form
ab
le 3
D
ob
j
ect
si
m
u
latio
n
can b
e
a g
o
o
d
can
d
i
d
a
te in
mo
b
ile
envi
ro
nm
ent
t
o
achi
e
ve
t
h
e
re
al
i
s
t
i
c
and
pl
ausi
bl
e
pe
r
f
o
rm
ance a
n
d
be
ha
vi
o
r
o
f
de
formable objects for thes
e
C
P
U a
n
d
G
P
U
paral
l
e
l
p
r
oces
si
ng
m
e
t
hods.
ACKNOWLE
DGE
M
ENTS
Th
is w
o
rk
w
a
s su
ppo
r
t
ed
b
y
th
e
Soo
n
c
hu
nhyan
g
Un
iv
er
sity
Resear
ch
Fun
d
.
REFERE
NC
ES
[1]
J
.
M
o
s
e
gaard, “
A
GP
U acceler
a
t
ed s
p
ring m
a
s
s
s
y
s
t
em
for s
u
rgi
cal s
i
m
u
la
tion”
,
Studies inHea
lth
Technology and
Informatics
, vo
l. 111, pp. 342–34
8, 2005
.
[2]
J.H. Jeon, M.H. Choi, Y.
S. Jeong and M. Hon
g
, “Hierarchical B
ounding Sphere FFD-AABB
Algorithm for Fast
Collision Handi
ng of 3D Deform
able Objec
t
s o
n
Sm
art Devices
”,
Journal of Internet Techno
log
y
, V
o
l 14
, N
o
. 5
,
pp. 843-850
, 20
13.
[3]
D. Baraff and W. Andrew, “L
arge
steps in cloth simulation”,
Proceedings of
the 25th annual conferen
ce on
Computer graphics and
inte
ra
ctive techniques.
ACM
, pp
.43-54, 1
998.
[4]
D. Ginsburg, B
.
Purnomo, D. Shreiner
,
and A.
Munshi, “Open
GL ES 3
.
0 Progr
amming Guide,” 2014.
[5]
S
.
Kazuki, and T. F
u
rum
o
to. “Grand centra
l
di
s
p
atch”
,
Pro Multithread
ing and Memory Manag
ement for iOS a
n
d
OS X.
Apress
, 13
9-145, 2012
.
[6]
V. Nahavand
ipo
o
r, “
C
oncurrent
Progr
am
m
i
ng in Mac OS X and iOS: Unleash Multicor
e Perform
ance wi
th Grand
Central Dispatch
”, O
'
Re
illy
Media, In
c., 2011
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Imp
l
emen
ta
tion
o
f
Clo
t
h S
i
mu
la
tion
Using
Pa
ra
llel Co
mpu
tin
g on
Mob
ile Device
(Min
H
ong
)
56
8
BIOGRAP
HI
ES OF
AUTH
ORS
Jae Hong Jeon
received BS d
e
gree
in Depar
tment of Compu
t
er Software
En
gineer
ing from
Soonchunh
y
a
ng
University
in
2012. Now he
is
undertak
ing
a master d
e
gree of
computer
engineering
cou
r
se as a member of th
e Com
puter Graphics
Lab
at Soonchunh
y
a
ng University
.
His
res
earch
int
e
res
t
s
ar
e com
p
uter g
a
m
e
deve
l
opm
ent, com
put
er graph
i
cs
, AR
(Augm
ented
Reality
) and embedded
motion
capture
Se
Dong M
i
n
was born in Seo
u
l, Korea,
in 19
75. He
r
e
c
e
ived
the M
.
S
.
and
P
h
.D. degr
ees
in
ele
c
tri
cal
and ele
c
troni
c engi
neering from
th
e Department of Electr
i
cal
and Electronics
Engineering, Y
onsei University
, Seoul, in 2004
and 2010, respectively
.
He is currently
an
Assistant Professor at the Dep
a
r
t
ment of Me
dical IT
Engineerin
g, Soonchunh
y
a
ng University
,
As
an, Korea.
His
res
earch ar
ea inc
l
udes
bio
m
edical s
i
gn
al
proces
s
i
ng, he
al
thcar
e s
e
ns
or
application, and
mobile health
car
e technolog
ies.
M
i
n Hong
is
an
As
s
o
ciate
P
r
ofe
s
s
o
r at th
e Depa
rtm
e
nt of Comp
uter Softwar
e
an
d Engineering
,
Soonchunh
y
a
ng
University
in
Asan, Korea. He
received
BS in Comput
er Scien
ce fro
m
Soonchunh
y
a
ng
University
in
1995. He also rece
ived MS in
Computer Science and PhD in
Bioinformatics from the University
of Colora
do
in 2001 and 2005, respectively
.
His research
inter
e
s
t
s
are in
Com
puter Graphics
, M
obi
le
Computing, Ph
y
s
ically
-
b
ased
Modeling and
Sim
u
lation, Bio
i
nform
a
tics Applic
ations
, and u
-
Health
car
e Applic
a
tions. In pr
esent, he is a
Director
of Com
puter Graph
i
cs
Laborator
y
at Soo
n
chunh
y
a
ng University
.
Evaluation Warning : The document was created with Spire.PDF for Python.