Skip to content
Projects
Groups
Snippets
Help
Loading...
Sign in / Register
Toggle navigation
O
opencv
Project
Project
Details
Activity
Cycle Analytics
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Charts
Issues
0
Issues
0
List
Board
Labels
Milestones
Merge Requests
0
Merge Requests
0
CI / CD
CI / CD
Pipelines
Jobs
Schedules
Charts
Packages
Packages
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Charts
Create a new issue
Jobs
Commits
Issue Boards
Open sidebar
submodule
opencv
Commits
87f4b47a
Commit
87f4b47a
authored
Jun 17, 2014
by
Ilya Lavrenov
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
optimized INTER_LINEAR mode
parent
c41a1343
Show whitespace changes
Inline
Side-by-side
Showing
3 changed files
with
60 additions
and
8 deletions
+60
-8
imgwarp.cpp
modules/imgproc/src/imgwarp.cpp
+2
-3
remap.cl
modules/imgproc/src/opencl/remap.cl
+57
-4
test_warp.cpp
modules/imgproc/test/ocl/test_warp.cpp
+1
-1
No files found.
modules/imgproc/src/imgwarp.cpp
View file @
87f4b47a
...
...
@@ -3640,10 +3640,9 @@ static bool ocl_remap(InputArray _src, OutputArray _dst, InputArray _map1, Input
}
int
scalarcn
=
cn
==
3
?
4
:
cn
;
int
sctype
=
CV_MAKETYPE
(
depth
,
scalarcn
);
buildOptions
+=
format
(
" -D T=%s -D T1=%s"
" -D cn=%d -D ST=%s"
,
buildOptions
+=
format
(
" -D T=%s -D T1=%s -D cn=%d -D ST=%s -D depth=%d"
,
ocl
::
typeToStr
(
type
),
ocl
::
typeToStr
(
depth
),
cn
,
ocl
::
typeToStr
(
sctype
));
cn
,
ocl
::
typeToStr
(
sctype
)
,
depth
);
ocl
::
Kernel
k
(
kernelName
.
c_str
(),
ocl
::
imgproc
::
remap_oclsrc
,
buildOptions
);
...
...
modules/imgproc/src/opencl/remap.cl
View file @
87f4b47a
...
...
@@ -274,7 +274,7 @@ __kernel void remap_16SC2_16UC1(__global const uchar * srcptr, int src_step, int
ST
nVal
)
{
int
x
=
get_global_id
(
0
)
;
int
y
=
get_global_id
(
1
)
;
int
y
=
get_global_id
(
1
)
*
rowsPerWI
;
if
(
x
<
dst_cols
)
{
...
...
@@ -313,7 +313,15 @@ __kernel void remap_16SC2_16UC1(__global const uchar * srcptr, int src_step, int
}
}
#
elif
INTER_LINEAR
#
elif
defined
INTER_LINEAR
__constant
float
coeffs[64]
=
{
1.000000f,
0.000000f,
0.968750f,
0.031250f,
0.937500f,
0.062500f,
0.906250f,
0.093750f,
0.875000f,
0.125000f,
0.843750f,
0.156250f,
0.812500f,
0.187500f,
0.781250f,
0.218750f,
0.750000f,
0.250000f,
0.718750f,
0.281250f,
0.687500f,
0.312500f,
0.656250f,
0.343750f,
0.625000f,
0.375000f,
0.593750f,
0.406250f,
0.562500f,
0.437500f,
0.531250f,
0.468750f,
0.500000f,
0.500000f,
0.468750f,
0.531250f,
0.437500f,
0.562500f,
0.406250f,
0.593750f,
0.375000f,
0.625000f,
0.343750f,
0.656250f,
0.312500f,
0.687500f,
0.281250f,
0.718750f,
0.250000f,
0.750000f,
0.218750f,
0.781250f,
0.187500f,
0.812500f,
0.156250f,
0.843750f,
0.125000f,
0.875000f,
0.093750f,
0.906250f,
0.062500f,
0.937500f,
0.031250f,
0.968750f
}
;
__kernel
void
remap_16SC2_16UC1
(
__global
const
uchar
*
srcptr,
int
src_step,
int
src_offset,
int
src_rows,
int
src_cols,
__global
uchar
*
dstptr,
int
dst_step,
int
dst_offset,
int
dst_rows,
int
dst_cols,
...
...
@@ -326,6 +334,7 @@ __kernel void remap_16SC2_16UC1(__global const uchar * srcptr, int src_step, int
if
(
x
<
dst_cols
)
{
WT
scalar
=
convertToWT
(
convertScalar
(
nVal
))
;
int
dst_index
=
mad24
(
y,
dst_step,
mad24
(
x,
TSIZE,
dst_offset
))
;
int
map1_index
=
mad24
(
y,
map1_step,
mad24
(
x,
(
int
)
sizeof
(
short2
)
,
map1_offset
))
;
int
map2_index
=
mad24
(
y,
map2_step,
mad24
(
x,
(
int
)
sizeof
(
ushort
)
,
map2_offset
))
;
...
...
@@ -347,7 +356,6 @@ __kernel void remap_16SC2_16UC1(__global const uchar * srcptr, int src_step, int
ushort
map2Value
=
(
ushort
)(
map2[0]
&
(
INTER_TAB_SIZE2
-
1
))
;
WT2
u
=
(
WT2
)(
map2Value
&
(
INTER_TAB_SIZE
-
1
)
,
map2Value
>>
INTER_BITS
)
/
(
WT2
)(
INTER_TAB_SIZE
)
;
WT
scalar
=
convertToWT
(
convertScalar
(
nVal
))
;
WT
a
=
scalar,
b
=
scalar,
c
=
scalar,
d
=
scalar
;
if
(
!NEED_EXTRAPOLATION
(
map_dataA.x,
map_dataA.y
))
...
...
@@ -390,6 +398,7 @@ __kernel void remap_2_32FC1(__global const uchar * srcptr, int src_step, int src
if
(
x
<
dst_cols
)
{
WT
scalar
=
convertToWT
(
convertScalar
(
nVal
))
;
int
dst_index
=
mad24
(
y,
dst_step,
mad24
(
x,
TSIZE,
dst_offset
))
;
int
map1_index
=
mad24
(
y,
map1_step,
mad24
(
x,
(
int
)
sizeof
(
float
)
,
map1_offset
))
;
int
map2_index
=
mad24
(
y,
map2_step,
mad24
(
x,
(
int
)
sizeof
(
float
)
,
map2_offset
))
;
...
...
@@ -403,6 +412,49 @@ __kernel void remap_2_32FC1(__global const uchar * srcptr, int src_step, int src
__global
const
float
*
map2
=
(
__global
const
float
*
)(
map2ptr
+
map2_index
)
;
__global
T
*
dst
=
(
__global
T
*
)(
dstptr
+
dst_index
)
;
#
if
defined
BORDER_CONSTANT
float
xf
=
map1[0],
yf
=
map2[0]
;
int
sx
=
convert_int_sat_rtn
(
xf
)
,
sy
=
convert_int_sat_rtn
(
yf
)
;
__constant
float
*
coeffs_x
=
coeffs
+
((
convert_int_rte
(
xf
*
INTER_TAB_SIZE
)
&
(
INTER_TAB_SIZE
-
1
))
<<
1
)
;
__constant
float
*
coeffs_y
=
coeffs
+
((
convert_int_rte
(
yf
*
INTER_TAB_SIZE
)
&
(
INTER_TAB_SIZE
-
1
))
<<
1
)
;
WT
sum
=
(
WT
)(
0
)
,
xsum
;
int
src_index
=
mad24
(
sy,
src_step,
mad24
(
sx,
TSIZE,
src_offset
))
;
#
pragma
unroll
for
(
int
yp
=
0
; yp < 2; ++yp, src_index += src_step)
{
if
(
sy
+
yp
>=
0
&&
sy
+
yp
<
src_rows
)
{
xsum
=
(
WT
)(
0
)
;
if
(
sx
>=
0
&&
sx
+
2
<
src_cols
)
{
#
if
depth
==
0
&&
cn
==
1
uchar2
value
=
vload2
(
0
,
srcptr
+
src_index
)
;
xsum
=
dot
(
convert_float2
(
value
)
,
(
float2
)(
coeffs_x[0],
coeffs_x[1]
))
;
#
else
#
pragma
unroll
for
(
int
xp
=
0
; xp < 2; ++xp)
xsum
=
fma
(
convertToWT
(
loadpix
(
srcptr
+
mad24
(
xp,
TSIZE,
src_index
)))
,
coeffs_x[xp],
xsum
)
;
#
endif
}
else
{
#
pragma
unroll
for
(
int
xp
=
0
; xp < 2; ++xp)
xsum
=
fma
(
sx
+
xp
>=
0
&&
sx
+
xp
<
src_cols
?
convertToWT
(
loadpix
(
srcptr
+
mad24
(
xp,
TSIZE,
src_index
)))
:
scalar,
coeffs_x[xp],
xsum
)
;
}
sum
=
fma
(
xsum,
coeffs_y[yp],
sum
)
;
}
else
sum
=
fma
(
scalar,
coeffs_y[yp],
sum
)
;
}
storepix
(
convertToT
(
sum
)
,
dst
)
;
#
else
float2
map_data
=
(
float2
)(
map1[0],
map2[0]
)
;
int2
map_dataA
=
convert_int2_sat_rtn
(
map_data
)
;
...
...
@@ -440,6 +492,7 @@ __kernel void remap_2_32FC1(__global const uchar * srcptr, int src_step, int src
c
*
(
1
-
u.x
)
*
(
u.y
)
+
d
*
(
u.x
)
*
(
u.y
)
;
storepix
(
convertToT
(
dst_data
)
,
dst
)
;
#
endif
}
}
}
...
...
@@ -454,6 +507,7 @@ __kernel void remap_32FC2(__global const uchar * srcptr, int src_step, int src_o
if
(
x
<
dst_cols
)
{
WT
scalar
=
convertToWT
(
convertScalar
(
nVal
))
;
int
dst_index
=
mad24
(
y,
dst_step,
mad24
(
x,
TSIZE,
dst_offset
))
;
int
map_index
=
mad24
(
y,
map_step,
mad24
(
x,
(
int
)
sizeof
(
float2
)
,
map_offset
))
;
...
...
@@ -473,7 +527,6 @@ __kernel void remap_32FC2(__global const uchar * srcptr, int src_step, int src_o
float2
_u
=
map_data
-
convert_float2
(
map_dataA
)
;
WT2
u
=
convertToWT2
(
convert_int2_rte
(
convertToWT2
(
_u
)
*
(
WT2
)
INTER_TAB_SIZE
))
/
(
WT2
)
INTER_TAB_SIZE
;
WT
scalar
=
convertToWT
(
convertScalar
(
nVal
))
;
WT
a
=
scalar,
b
=
scalar,
c
=
scalar,
d
=
scalar
;
if
(
!NEED_EXTRAPOLATION
(
map_dataA.x,
map_dataA.y
))
...
...
modules/imgproc/test/ocl/test_warp.cpp
View file @
87f4b47a
...
...
@@ -267,7 +267,7 @@ PARAM_TEST_CASE(Remap, MatDepth, Channels, std::pair<MatType, MatType>, BorderTy
Border
map1Border
=
randomBorder
(
0
,
useRoi
?
MAX_VALUE
:
0
);
randomSubMat
(
map1
,
map1_roi
,
dstROISize
,
map1Border
,
map1Type
,
-
mapMaxValue
,
mapMaxValue
);
Border
map2Border
=
randomBorder
(
0
,
useRoi
?
MAX_VALUE
:
0
);
Border
map2Border
=
randomBorder
(
0
,
useRoi
?
MAX_VALUE
+
1
:
0
);
if
(
map2Type
!=
noType
)
{
int
mapMinValue
=
-
mapMaxValue
;
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment