1. INTRODUCTION
TO
BIDIRECTIONAL
PATH
TRACING
&
ITS
IMPLEMENTATION
USING
OPENCL
双方向パストレーシング(BDPT)の基礎からOPENCLによる実装まで
TAKAHIRO
HARADA
(AMD)
SHO
IKEDA
(RICOH)
SYOYO
FUJITA
(LTE)

2. 2
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
OVERVIEW
OF
THIS
TALK
y IntroducJon
(Harada,
20min
-­‐
25min)
‒ Start
with
path
tracing
review
‒ Transforming
path
tracing
to
bidirecJonal
path
tracing
y BidirecJonal
Path
Tracing
(Ikeda,
20min
-­‐
25min)
‒ Classic
BDPT
‒ Instant
BDPT
‒ Lvc
BDPT
y IntegraJon
of
BidirecJonal
Path
Tracing
to
the
Engine
(Harada,
5min)
y Examples
(Harada,
5min)

3. 3
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
(MONTE
CARLO
RAY
TRACING)
y !=
Raster
graphics
(game)
y BeauJful
rendering
y Global
illuminaJon

4. 4
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
(MONTE
CARLO
RAY
TRACING)
y ComputaJonally
expensive
‒ Take
Jme
to
converge
‒ Noisy
at
first,
gebng
cleaner
later
y Make
1
step
rendering
faster
‒ Total
Jme
decreases
‒ OpJmizaJon
(using
more
CPU
cores,
GPUs)
‒ OpenCL
(my
talk@CEDEC2014)
y Make
1
step
less
noisy
‒ Need
less
steps
‒ Beeer
sampling
techniques
‒ Clever
algorithm
‒ BDPT
(this
talk)

5. PATH
TRACING
IMPORTANCE
SAMPLING
MULTIPLE
IMPORTANCE
SAMPLING

6. 6
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
FIRST
PATH
TRACING

7. 7
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
IMPLEMENTATION
#1
for(int i=0; i<nx; i++) for(int j=0; j<ny; j++)!
{!
Ray ray = genPrimaryRay( i, j );!
!
float4 coeff = 1.f;!
float4& output = pixel[i,j];!
for(int depth=0; depth<maxDepth; depth++)!
{!
Hit hit = intersect( ray );!
if( !hit.hasHit() ) break;!
!
if( hit.isEmissive() )!
{// Implicit Connection!
output += coeff * getEmission( hit );!
break;!
}!
!
ray, f, pdf = randomSample( hit );!
coeff *= f * dot( hit.m_n, ray.m_dir ) / pdf;!
}!
}!

8. 8
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
SAMPLING
NEXT
RAY
y For
non
transmission
surface,
no
need
to
sample
the
other
side
RANDOM
SAMPLING
for(int i=0; i<nx; i++) for(int j=0; j<ny; j++)!
{!
Ray ray = genPrimaryRay( i, j );!
!
float4 coeff = 1.f;!
float4& output = pixel[i,j];!
for(int depth=0; depth<maxDepth; depth++)!
{!
Hit hit = intersect( ray );!
if( !hit.hasHit() ) break;!
!
if( hit.isEmissive() )!
{// Implicit Connection!
output += coeff * getEmission( hit );!
break;!
}!
!
ray, f, pdf = randomSample( hit );!
coeff *= f * dot( hit.m_n, ray.m_dir ) / pdf;!
}!
}!

9. 9
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
SAMPLING
NEXT
RAY
y Random
sampling
is
not
always
efficient
‒ Diffuse
surface
=>
OK
‒ Glossy
surface
=>
Hmm
L
‒ Specular
surface
=>
Bad
L
y We
usually
know
the
reflecJon
characterisJcs
about
surface
‒ Why
not
use
them?
‒ “Importance
sampling”
(BRDF)
RANDOM
SAMPLING

10. 10
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
IMPORTANCE
SAMPLING
(BRDF)
y Specular
surface
‒ Only
reflect
in
the
mirrored
direcJon
‒ Always
sample
the
direcJon
(pdf
=
1)
y Glossy
surface
‒ Mostly
reflect
around
mirrored
direcJon
‒ Sample
around
the
mirrored
direcJon
Specular
(R)
Glossy
Mae
Specular
(T)

11. 11
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
IMPLEMENTATION
#2
for(int i=0; i<nx; i++) for(int j=0; j<ny; j++)!
{!
Ray ray = genPrimaryRay( i, j );!
!
float4 coeff = 1.f;!
float4& output = pixel[i,j];!
for(int depth=0; depth<maxDepth; depth++)!
{!
Hit hit = intersect( ray );!
if( !hit.hasHit() ) break;!
!
if( hit.isEmissive() )!
{// Implicit Connection!
output += coeff * getEmission( hit );!
break;!
}!
!
ray, f, pdf = randomSample( hit );!
coeff *= f * dot( hit.m_n, ray.m_dir ) / pdf;!
}!
}!
for(int i=0; i<nx; i++) for(int j=0; j<ny; j++)!
{!
Ray ray = genPrimaryRay( i, j );!
!
float4 coeff = 1.f;!
float4& output = pixel[i,j];!
for(int depth=0; depth<maxDepth; depth++)!
{!
Hit hit = intersect( ray );!
if( !hit.hasHit() ) break;!
!
if( hit.isEmissive() )!
{// Implicit Connection!
output += coeff * getEmission( hit );!
break;!
}!
!
ray, f, pdf = brdfSample( hit );!
coeff *= f * dot( hit.m_n, ray.m_dir ) / pdf;!
}!
}!
Random
Sampling
Brdf
Importance
Sampling

12. 12
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
IMPLEMENTATION
#2
for(int i=0; i<nx; i++) for(int j=0; j<ny; j++)!
{!
Ray ray = genPrimaryRay( i, j );!
!
float4 coeff = 1.f;!
float4& output = pixel[i,j];!
for(int depth=0; depth<maxDepth; depth++)!
{!
Hit hit = intersect( ray );!
if( !hit.hasHit() ) break;!
!
if( hit.isEmissive() )!
{// Implicit Connection!
output += coeff * getEmission( hit );!
break;!
}!
!
ray, f, pdf = randomSample( hit );!
coeff *= f * dot( hit.m_n, ray.m_dir ) / pdf;!
}!
}!
for(int i=0; i<nx; i++) for(int j=0; j<ny; j++)!
{!
Ray ray = genPrimaryRay( i, j );!
!
float4 coeff = 1.f;!
float4& output = pixel[i,j];!
for(int depth=0; depth<maxDepth; depth++)!
{!
Hit hit = intersect( ray );!
if( !hit.hasHit() ) break;!
!
if( hit.isEmissive() )!
{// Implicit Connection!
output += coeff * getEmission( hit );!
break;!
}!
!
ray, f, pdf = brdfSample( hit );!
coeff *= f * dot( hit.m_n, ray.m_dir ) / pdf;!
}!
}!
Random
Sampling
Brdf
Importance
Sampling

13. 13
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
INEFFICIENT
CASE
y Rough
surface
+
Small
light
y Brdf
sampling
rarely
hits
the
light
y SoluJon
‒ We
know
where
the
lights
are
‒ Why
ignoring??
‒ Generate
ray
from
lights
‒ “Importance
sampling”
(Light)
Rough
Reference
Brdf
sampling

14. 14
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
INEFFICIENT
CASE
y Rough
surface
+
Small
light
y Brdf
sampling
rarely
hits
the
light
y SoluJon
‒ We
know
where
the
lights
are
‒ Why
ignoring??
‒ Generate
ray
from
lights
‒ “Importance
sampling”
(Light)
Rough
Reference
Brdf
sampling
Light
sampling

15. 15
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
IMPLEMENTATION
#3
for(int i=0; i<nx; i++) for(int j=0; j<ny; j++)!
{!
Ray ray = genPrimaryRay( i, j );!
!
float4 coeff = 1.f;!
float4& output = pixel[i,j];!
for(int depth=0; depth<maxDepth; depth++)!
{!
Hit hit = intersect( ray );!
if( !hit.hasHit() ) break;!
!
if( hit.isEmissive() )!
{// Implicit Connection!
output += coeff * getEmission( hit );!
break;!
}!
!
ray, f, pdf = brdfSample( hit );!
coeff *= f * dot( hit.m_n, ray.m_dir ) / pdf;!
}!
}!
for(int i=0; i<nx; i++) for(int j=0; j<ny; j++)!
{!
Ray ray = genPrimaryRay( i, j );!
!
float4 coeff = 1.f;!
float4& output = pixel[i,j];!
for(int depth=0; depth<maxDepth; depth++)!
{!
Hit hit = intersect( ray );!
if( !hit.hasHit() ) break;!
{// Explicit Connection!
ray, pdf = lightSample( hit );!
Hit hit1 = intersect( ray );!
if( !hit1.hasHit() )!
{!
output+=coeff*getEmission( hit1 )/pdf;!
}!
}!
ray, f, pdf = brdfSample( hit );!
coeff *= f * dot( hit.m_n, ray.m_dir ) / pdf;!
}!
}!
Brdf
Importance
Sampling
Light
Importance
Sampling

16. 16
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
IMPLICIT
CONNECTION,
EXPLICIT
CONNECTION
y Implicit
ConnecJon
‒ A
ray
accidentally
hit
a
light
source
‒ Ater
primary
rays
are
generated
from
camera
‒ Ater
bounced
rays
are
sampled
from
BRDF
y Explicit
ConnecJon
‒ A
path
is
intenJonally
connected
to
a
light
source
‒ Direct
illuminaJon
(sample
a
vertex
on
a
light
source)
Camera
Light
Object
Camera
Light
Object

17. 17
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
IMPLEMENTATION
#3
y We
should
not
accumulaJng
implicit
hits
for
this
implementaBon
y Even
if
a
ray
hits
an
emissive
surface,
we
ignore
it
y Sounds
like
we
are
wasJng
something
NOTE

18. 18
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
BRDF
SAMPLING
+
LIGHT
SAMPLING
y Both
generate
image
‒ Each
technique
has
strong
&
weak
points
‒ Converges
to
the
same
image
at
the
end
Brdf
sampling
(16spp)
Light
sampling
(16spp)
Brdf
sampling
(inf. spp)
Light
sampling
(inf. spp)
=

19. 19
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
IMPLEMENTATION
#3
y We
should
not
accumulaJng
implicit
hits
for
this
implementaBon
y Even
if
a
ray
hits
an
emissive
surface,
we
ignore
it
y Sounds
like
we
are
wasJng
something
y Can
we
use
it
somehow?
NOTE

20. 20
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
BRDF
SAMPLING
+
LIGHT
SAMPLING
y Then
simply…
Take
an
average?
‒ Works
‒ but
not
the
best
y We
can
set
any
coefficients
if
they
sum
up
to
1.0
‒ brdfSampledImage
*
0.2
+
lightSampledImage
*
0.8
?
‒ brdfSampledImage
*
0.8
+
lightSampledImage
*
0.2
?
‒ Hmm…
x
0.5
+
x
0.5
Brdf
sample
Light
sample
x
0.2
+
x
0.8
Brdf
sample
Light
sample

21. 21
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
J
L
COMPARING
BRDF
SAMPLING
AND
LIGHT
SAMPLING
y Light
sampling
‒ Good
for
rough
surface
‒ Bad
for
sharp
surface
y Brdf
sampling
‒ Bad
for
rough
surface
‒ Good
for
sharp
surface
L
J

22. 22
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
BRDF
SAMPLING
+
LIGHT
SAMPLING
y Then
simply…
Take
an
average?
‒ Works
‒ but
not
the
best
y We
can
set
any
coefficients
if
they
sum
up
to
1.0
‒ brdfSampledImage
*
0.2
+
lightSampledImage
*
0.8
?
‒ brdfSampledImage
*
0.8
+
lightSampledImage
*
0.2
?
‒ Hmm
y We
do
not
have
to
se
the
same
coefficients
for
all
the
pixels
J
‒ “MulJple
importance
sampling”
‒ Per
pixel
weight
x
0.5
+
x
0.5
Brdf
sample
Light
sample
x
0.2
+
x
0.8
Brdf
sample
Light
sample
x
+
x
=
Brdf
sample
Light
sample
w_b
w_l
MIS

23. 23
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
BRDF
SAMPLING
+
LIGHT
SAMPLING
y Uniform
weighJng
(Average)
y Per
pixel
weighJng
(MulJple
Importance
Sampling,
MIS)
VISUALIZATION
OF
MIS
x
+
x
=
Brdf
sample
Light
sample
w_b
w_l
MIS
+
=
1.0
w_b
w_l
x
+
x
=
Brdf
sample
Light
sample
w_b
w_l
Average
0.5
0.5
0.5
0.5
+
=
1.0
w_b
w_l

24. 24
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
MULTIPLE
IMPORTANCE
SAMPLING
y Light
sampling
y Brdf
sampling
0.1
0.9
0.9
0.1
0.5
0.5

25. 25
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
MULTIPLE
IMPORTANCE
SAMPLING
WEIGHT
y Detailed
look
at
brdf
&
light
sampling
(direct
illuminaJon)
y Possible
to
sample
the
same
direcJon
using
both
sampling
technique
‒ Only
difference
is
the
PDF
(probability
density
funcJon)
y Rough
surface
‒ PDF
of
Brdf
sampling
is
almost
uniform
‒ Low
confidence
in
the
sample
=>
Higher
noise
‒ PDF
of
light
sampling
has
a
strong
spike
at
light
direcJon
‒ High
confidence
in
the
sample
=>
Lower
noise
Brdf
Sampling
Light
Sampling

26. 26
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
MULTIPLE
IMPORTANCE
SAMPLING
WEIGHT
y Detailed
look
at
brdf
&
light
sampling
(direct
illuminaJon)
y Possible
to
sample
the
same
direcJon
using
both
sampling
technique
‒ Only
difference
is
the
PDF
(probability
density
funcJon)
y Sharp
surface
‒ PDF
of
Brdf
sampling
is
almost
uniform
‒ Very
high
confidence
in
the
sample
=>
Lower
noise
‒ PDF
of
light
sampling
has
a
strong
spike
at
light
direcJon
‒ RelaJvely
lower
confidence
in
the
sample
=>
Higher
noise
Brdf
Sampling
Light
Sampling

27. 27
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
MULTIPLE
IMPORTANCE
SAMPLING
WEIGHT
y Use
sampling
PDF
to
compute
weight
y Weight
for
light
sample
(Explicit
hit)
y Weight
for
brdf
sample
(Implicit
hit)
y This
is
very
important
for
an
efficient
BDPT

28. 28
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
IMPLEMENTATION
#4
(FINAL)
for(int i=0; i<nx; i++) for(int j=0; j<ny; j++)!
{!
Ray ray = genPrimaryRay( i, j );!
!
float4 coeff = 1.f;!
float4& output = pixel[i,j];!
float pdfb = 0.f;!
for(int depth=0; depth<maxDepth; depth++)!
{!
Hit hit = intersect( ray );!
if( !hit.hasHit() ) break;!
if( hit.isEmissive() )!
{// Implicit Connection!
pdfl = lightPdf( hit );!
w = pdfb / (pdfb + pdfl);!
output += coeff * getEmission( hit ) * w;!
break;!
}!
{// Explicit Connection!
ray, pdfl = lightSample( hit );!
Hit hit1 = intersect( ray );!
if( !hit1.hasHit() )!
{!
w = pdfl / (pdfb + pdfl);!
output += coeff * getEmission( hit1 ) * w / pdfl;!
}!
}!
ray, f, pdfb = brdfSample( hit );!
coeff *= f * dot( hit.m_n, nextRay.m_dir ) / pdfb;!
}!
}!

29. BEYOND
PATH
TRACING

30. 30
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
OUTDOOR

31. 31
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
INDOOR/OUTDOOR

32. 32
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
INDOOR

33. 33
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
IS
NOT
PERFECT
y Path
tracing
is
really
good
at
an out
door
scene
y But
not
for
a
scene
mostly
lit
by
indirect
illuminaJon
(indoor,
day
Jme)

34. 34
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
IS
NOT
PERFECT
64
spp
256
spp
1024
spp
x
4
x
4
Indoor
Indoor/outdoor
Outdoor
Good
Ok
Bad
Path
Tracing
is..

35. 35
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
BAD
CONVERGENCE
FOR
INDOOR
SCENE
y Ray
needs
to
be
bounced
more
than
once
before
connecJng
to
the
light
source
y As
we
bounce,
we
get
less
&
less
confident
about
the
path
‒ Lower
probability
=>
More
noise

36. 36
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
LIGHT
TRACING
y If
traced
from
the
light,
it
is
beeer
(more
confident)
y Path
Tracing
y Light
Tracing
Camera
Light
Object
Camera
Light
Object
Path
Tracing
BDPT

37. 37
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
+
1
LIGHT
PATH
y Can
we
mix
path
tracing
&
light
tracing
‒ Trace
1
light
path
‒ Connect
to
camera
path
Camera
Light
Object

38. 38
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
+
1
LIGHT
PATH
y Can
we
mix
path
tracing
&
light
tracing
‒ Trace
1
light
path
‒ Connect
to
camera
path
y Improves
some
situaJons
(not
all)
‒ Example.
A
room
is
lit
by
bounced
light
y Can
we
trace
more
than
1
light
path,
and
combine?
‒ =>
BidirecJonal
Path
Tracing
Camera
Light
Object
Path
Tracing
Path
Trace
+
1
light
path
64
spp

39. BDPT

40. Bi-Directional Path Tracing (BDPT)
Camera path
Light path
Connection
Camera
Light
Object
• BDPT traces paths from the camera and the light source
• BDPT combines various path sampling strategies

41. Bi-Directional Path Tracing (BDPT)
Camera path
Light path
Connection
Camera
Light
Object
• BDPT traces paths from the camera and the light source
• BDPT combines various path sampling strategies

42. Bi-Directional Path Tracing (BDPT)
Camera path
Light path
Connection
Camera
Light
Object
• BDPT traces paths from the camera and the light source
• BDPT combines various path sampling strategies

43. Bi-Directional Path Tracing (BDPT)
Camera path
Light path
Connection
Camera
Light
Object
• BDPT traces paths from the camera and the light source
• BDPT combines various path sampling strategies

44. BDPT
I = w0
w1
+
+
…
wk
·
·
·
x0 xk
• The contributions of the sampling strategies are weighted

45. BDPT + Multiple Importance Sampling (MIS)
Xs = x0, · · · xs + xs+1, · · · xk
x0 xk
xs
xs+1
• MIS weight computation
ws(X) =
ps(X)
Pk
i=0 pi(X)

46. x0 xk
xs
xs+1
Xs = x0, · · · xs + xs+1, · · · xk
ps(X) = !p 0 · !p 1 · · · !p s · p s+1 · · · p k 1 · p k
BDPT + Multiple Importance Sampling (MIS) (2
The product of the pdfs of the path sampling

47. The problem of BDPT on GPU implementation
0 1 2 3Global Memory
0 1 2
• BDPT stores all the vertices on the camera and light paths
• The global memory consumption is high

48. 0 1 2 3Global Memory
0 1 2
• BDPT stores all the vertices on the camera and light paths
• The global memory consumption is high
The problem of BDPT on GPU implementation

49. 0 1 2 3Global Memory
0 1 2
• BDPT stores all the vertices on the camera and light paths
• The global memory consumption is high
The problem of BDPT on GPU implementation

50. 0 1 2 3Global Memory
0 1 2
• BDPT stores all the vertices on the camera and light paths
• The global memory consumption is high
The problem of BDPT on GPU implementation

51. p Instant BDPT
p Lvc BDPT
BDPT algorithm with low memory consumption

52. Instant BDPT
Implicit view ( ) path
Explicit view ( ) path
Implicit light ( ) path
Explicit light ( ) path
VI
VE LE
LI
• Path sampling strategies are limited to 4 (PT and LT)
• Low memory consumption
• If the camera has no collision detection, 3 sampling strategies

53. Instant BDPT + MIS
Explicit view ( ) path
VE
wVE
(X) =
pVE
(X)
pVE
(X) + pVI
(X) + pLE
(X)
1
wVE
(X)
= 1 +
pVI
(X)
pVE
(X)
+
pLE
(X)
pVE
(X)

54. Instant BDPT + MIS (2
1
wVE
(X)
= 1 +
pVI
(X)
pVE
(X)
+
pLE
(X)
pVE
(X)
pVI
(X)
pVE
(X)
=
!p k 1
p k+1
!p k 1
p k+1

55. Instant BDPT + MIS (3
1
wVE
(X)
= 1 +
pVI
(X)
pVE
(X)
+
pLE
(X)
pVE
(X)
x0
xk
s0 =
1
P(x 1 ! x0 ! x1)
pLE
(X)
pVE
(X)
= s0s1 · · · sk
x1

56. Instant BDPT + MIS (3
1
wVE
(X)
= 1 +
pVI
(X)
pVE
(X)
+
pLE
(X)
pVE
(X)
x0
xk
s1 =
1
P(x0 ! x1 ! x2)G(x0 $ x1)
pLE
(X)
pVE
(X)
= s0s1 · · · sk
x1

57. Instant BDPT + MIS (3
1
wVE
(X)
= 1 +
pVI
(X)
pVE
(X)
+
pLE
(X)
pVE
(X)
x0
xk
pLE
(X)
pVE
(X)
= s0s1 · · · sk sk = P(xk 1 xk xk+1)G(xk 1 $ xk)
x1

58. Instant BDPT OpenCL implementation

59. Instant BDPT : Bad case1
x0 xk

60. x0 xk
Instant BDPT : Bad case1

61. x0 xk
Instant BDPT : Bad case1

62. x0 xk
Instant BDPT : Bad case1

63. x0
xk
Specular
Instant BDPT : Bad case

64. x0
xk
Specular
Instant BDPT : Bad case

65. x0
xk
Specular
Instant BDPT : Bad case

66. x0
xk
Specular
Instant BDPT : Bad case

67. p Instant BDPT
p Lvc BDPT
BDPT algorithm with low memory consumption

68. Lvc BDPT light vertex cache
0 1 2 3Light Vertex Cache
• Light vertex cache
• Recursive MIS

69. Lvc BDPT light vertex cache
0 1 2 3Light Vertex Cache
• Light vertex cache
• Recursive MIS

70. Lvc BDPT light vertex cache
0 1 2 3Light Vertex Cache
• Light vertex cache
• Recursive MIS

71. Lvc BDPT light vertex cache
0 1 2 3Light Vertex Cache
• Light vertex cache
• Recursive MIS
• LvcBDPT doesn’t have the global memory for the camera path

72. Recursive MIS
x0
xk
xs
ws(X) =
ps(X)
Pk
i=0 pi(X)
1
ws(X)
=
Pk
i=0 pi(X)
ps(X)
p s+1 · dE
s + 1 + !p s · dL
s+1=

73. 1
ws(X)
= p s+1 · dE
s + 1 + !p s · dL
s+1
dE
1 =
1
!p 0
dE
s =
1 + p sdE
s 1
!p s 1
dL
k =
1
p k+1
Recursive MIS
x0
xk
xs
dL
s+1 =
1 + !p s+1 · dL
s+2
p s+2

74. 1
ws(X)
= p s+1 · dE
s + 1 + !p s · dL
s+1
Lvc BDPT + Recursive MIS
x0 xk
xs
dE
1 =
1
!p 0

75. 1
ws(X)
= p s+1 · dE
s + 1 + !p s · dL
s+1
x0 xk
xs
dE
s =
1 + p sdE
s 1
!p s 1
Lvc BDPT + Recursive MIS

76. 1
ws(X)
= p s+1 · dE
s + 1 + !p s · dL
s+1
x0 xk
xs
dE
s =
1 + p sdE
s 1
!p s 1
dL
k =
1
p k+1
Lvc BDPT + Recursive MIS

77. 1
ws(X)
= p s+1 · dE
s + 1 + !p s · dL
s+1
x0 xk
xs
dE
s =
1 + p sdE
s 1
!p s 1
dL
k 1 =
1 + !p k 1 · dL
k
p k
Lvc BDPT + Recursive MIS

78. 1
ws(X)
= p s+1 · dE
s + 1 + !p s · dL
s+1
x0 xk
xs
dE
s =
1 + p sdE
s 1
!p s 1
dL
s+1 =
1 + !p s+1 · dL
s+2
p s+2
Lvc BDPT + Recursive MIS

79. Lvc BDPT OpenCL implementation

80. InstantBDPT
LvcBDPT

81. InstantBDPT
LvcBDPT

82. InstantBDPT
LvcBDPT

83. INTEGRATION
TO
THE
ENGINE

84. 84
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
Back
End
World
GPU
(OCL)
-­‐ Path
Tracing
-­‐ Light
Tracing
-­‐
-­‐
World
CPU
-­‐ Path
Tracing
(OpJmized)
Front
End
ENGINE
ARCHITECTURE
OVERVIEW
Data
Loader
-­‐ Obj
-­‐ Eson
-­‐ Fbx
Data
Manager
-­‐ Texture
-­‐ Material
-­‐ Mesh
Tahoe
API
Texture
Loader
-­‐ Png
-­‐ Hdr
-­‐ OpenExr
-­‐ …
Factory
(CPU)
Camera
-­‐ PerspecJve
-­‐ Parallel
-­‐ Bake
-­‐ VR
Material
System
-­‐ Default
-­‐ Graph
-­‐ OSL
Light
Sampler
-­‐ Uniform
-­‐ Group
-­‐ StochasJc
Factory
(OCL)

85. 85
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
Back
End
World
GPU
(OCL)
-­‐ Path
Tracing
-­‐ Light
Tracing
-­‐ BidirecJonal
Path
Tracing
(LvcBDPT)
-­‐ BidirecJonal
Path
Tracing
(InstantBDPT)
World
CPU
-­‐ Path
Tracing
(OpJmized)
Front
End
ENGINE
ARCHITECTURE
OVERVIEW
Data
Loader
-­‐ Obj
-­‐ Eson
-­‐ Fbx
Data
Manager
-­‐ Texture
-­‐ Material
-­‐ Mesh
Tahoe
API
Texture
Loader
-­‐ Png
-­‐ Hdr
-­‐ OpenExr
-­‐ …
Factory
(CPU)
Camera
-­‐ PerspecJve
-­‐ Parallel
-­‐ Bake
-­‐ VR
Material
System
-­‐ Default
-­‐ Graph
-­‐ OSL
Light
Sampler
-­‐ Uniform
-­‐ Group
-­‐ StochasJc
Factory
(OCL)

86. 86
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
CODE
y Not
employing
a
mega
kernel
implementaJon
y There
are
many
small
kernels
for
path
tracing
‒ Easier
to
debug
‒ Performance
‒ Extendibility
=>
Key
to
add
BDPT
y (See
my
slides
@
CEDEC
2014)

87. 87
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
PATH
TRACING
KERNELS
y Prepare
‒ Camera
ray
set
up
y Loop
‒ Ray
cast
‒ Implicit
hit
accumulaJon
‒ Material
prepare
‒ Shadow
ray
set
up
‒ Material
evaluaJon
‒ Ray
cast
‒ Explicit
hit
accumulaJon
‒ Sample
next
ray
Path
Tracing
Each
line
is
one
or
more
OCL
kernels

88. 88
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
INSTANT
BDPT
y Prepare
‒ Set
up
camera
ray
‒ Sample
light
point
y Loop
‒ Ray
cast
‒ Material
prepare
(sampleBrdf)
‒ Implicit
connecJon
‒ Explicit
connecJon
to
light
‒ Make
shadow
ray
‒ Ray
cast
‒ Connect
to
light
‒ Sample
next
ray
‒ Update
MIS
weight
term
Trace
Light
Path
Trace
Camera
Path
Orange:
New
kernels
wrieen
for
BDPT
White
:
Using
path
tracing
kernels

89. 89
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
INSTANT
BDPT
y Prepare
‒ Set
up
light
ray
‒ Sample
lens
point
‒ Explicit
connecJon
light
to
camera
y Loop
‒ Ray
cast
‒ Material
prepare
(sampleBrdf)
‒ Explicit
connecJon
to
camera
‒ Make
shadow
ray
‒ Ray
cast
‒ Set
pixel
index
‒ Connect
to
camera
‒ Sample
next
ray
‒ Update
MIS
weight
term
y Prepare
‒ Set
up
camera
ray
‒ Sample
light
point
y Loop
‒ Ray
cast
‒ Material
prepare
(sampleBrdf)
‒ Implicit
connecJon
‒ Explicit
connecJon
to
light
‒ Make
shadow
ray
‒ Ray
cast
‒ Connect
to
light
‒ Sample
next
ray
‒ Update
MIS
weight
term
Trace
Light
Path
Trace
Camera
Path
Orange:
New
kernels
wrieen
for
BDPT
White
:
Using
path
tracing
kernels

90. 90
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
LVCBDPT
y More
complicated
than
Instant
BDPT
y ImplementaJon
is
not
that
far
from
Instant
BDPT

91. EXAMPLES

92. 92
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
BDPT
PT

93. 93
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
BDPT
PT

94. 94
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
BDPT
PT

95. 95
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
CLOSING
y References
‒ AddiJonal
note
‒ To
be
published
in
CEDEC
library
‒ Detailed
explanaJon
of
Instant
BDPT,
Lvc
BDPT
‒ CEDEC
2013
‒ hep://www.slideshare.net/takahiroharada/introducJon-­‐to-­‐monte-­‐carlo-­‐ray-­‐tracing-­‐cedec2013
‒ CEDEC
2014
‒ hep://www.slideshare.net/takahiroharada/introducJon-­‐to-­‐monte-­‐carlo-­‐ray-­‐tracing-­‐opencl-­‐implementaJon-­‐cedec-­‐2014
y Feedback
is
welcome
‒ For
next
years
presentaJon

96. 96
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
NON
SYMMETRIC
BRDF
Without
Fix
With
Fix

97. 97
|
OPENCL
BDPT
|
AUGUST
28,
2015
|
HARADA,
IKEDA,
FUJITA
NON
SYMMETRIC
BRDF
With
Fix
With
Fix