Skip to content
GitLab
Explore
Sign in
Primary navigation
Search or go to…
Project
G
GPUSort
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Wiki
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Snippets
Build
Pipelines
Jobs
Pipeline schedules
Artifacts
Deploy
Releases
Model registry
Operate
Environments
Monitor
Incidents
Analyze
Value stream analytics
Contributor analytics
CI/CD analytics
Repository analytics
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
Community forum
Contribute to GitLab
Provide feedback
Keyboard shortcuts
?
Snippets
Groups
Projects
Show more breadcrumbs
TNL
GPUSort
Commits
5617f462
There was an error fetching the commit references. Please try again later.
Commit
5617f462
authored
3 years ago
by
Xuan Thang Nguyen
Browse files
Options
Downloads
Patches
Plain Diff
fix copy and add support for block bitonic sort of bigger size
parent
e7519920
No related branches found
No related tags found
No related merge requests found
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
src/bitonicSort/bitonicSort.h
+36
-28
36 additions, 28 deletions
src/bitonicSort/bitonicSort.h
with
36 additions
and
28 deletions
src/bitonicSort/bitonicSort.h
+
36
−
28
View file @
5617f462
...
...
@@ -191,27 +191,32 @@ __device__ void bitonicSort_Block(TNL::Containers::ArrayView<Value, TNL::Devices
//------------------------------------------
//bitonic activity
{
int
i
=
threadIdx
.
x
;
int
paddedSize
=
closestPow2_ptx
(
src
.
getSize
());
for
(
int
monotonicSeqLen
=
2
;
monotonicSeqLen
<=
paddedSize
;
monotonicSeqLen
*=
2
)
{
//calculate the direction of swapping
int
monotonicSeqIdx
=
i
/
(
monotonicSeqLen
/
2
);
bool
ascending
=
(
monotonicSeqIdx
&
1
)
!=
0
;
if
((
monotonicSeqIdx
+
1
)
*
monotonicSeqLen
>=
src
.
getSize
())
//special case for parts with no "partner"
ascending
=
true
;
for
(
int
len
=
monotonicSeqLen
;
len
>
1
;
len
/=
2
)
{
//calculates which 2 indexes will be compared and swap
int
part
=
i
/
(
len
/
2
);
int
s
=
part
*
len
+
(
i
&
((
len
/
2
)
-
1
));
int
e
=
s
+
len
/
2
;
for
(
int
i
=
threadIdx
.
x
;
;
i
+=
blockDim
.
x
)
//simulates other blocks in case src.size > blockDim.x*2
{
//calculates which 2 indexes will be compared and swap
int
part
=
i
/
(
len
/
2
);
int
s
=
part
*
len
+
(
i
&
((
len
/
2
)
-
1
));
int
e
=
s
+
len
/
2
;
if
(
e
>=
src
.
getSize
())
//touching virtual padding, the order dont swap
break
;
//calculate the direction of swapping
int
monotonicSeqIdx
=
i
/
(
monotonicSeqLen
/
2
);
bool
ascending
=
(
monotonicSeqIdx
&
1
)
!=
0
;
if
((
monotonicSeqIdx
+
1
)
*
monotonicSeqLen
>=
src
.
getSize
())
//special case for parts with no "partner"
ascending
=
true
;
if
(
e
<
src
.
getSize
())
//not touching virtual padding
cmpSwap
(
sharedMem
[
s
],
sharedMem
[
e
],
ascending
,
Cmp
);
__syncthreads
();
}
__syncthreads
();
//only 1 synchronization needed
}
}
}
...
...
@@ -232,29 +237,32 @@ __device__ void bitonicSort_Block(TNL::Containers::ArrayView<Value, TNL::Devices
* */
template
<
typename
Value
,
typename
Function
>
__device__
void
bitonicSort_Block
(
TNL
::
Containers
::
ArrayView
<
Value
,
TNL
::
Devices
::
Cuda
>
src
,
TNL
::
Containers
::
ArrayView
<
Value
,
TNL
::
Devices
::
Cuda
>
dst
,
const
Function
&
Cmp
)
{
int
i
=
threadIdx
.
x
;
int
paddedSize
=
closestPow2_ptx
(
src
.
getSize
());
for
(
int
monotonicSeqLen
=
2
;
monotonicSeqLen
<=
paddedSize
;
monotonicSeqLen
*=
2
)
{
//calculate the direction of swapping
int
monotonicSeqIdx
=
i
/
(
monotonicSeqLen
/
2
);
bool
ascending
=
(
monotonicSeqIdx
&
1
)
!=
0
;
if
((
monotonicSeqIdx
+
1
)
*
monotonicSeqLen
>=
src
.
getSize
())
//special case for parts with no "partner"
ascending
=
true
;
for
(
int
len
=
monotonicSeqLen
;
len
>
1
;
len
/=
2
)
{
//calculates which 2 indexes will be compared and swap
int
part
=
i
/
(
len
/
2
);
int
s
=
part
*
len
+
(
i
&
((
len
/
2
)
-
1
));
int
e
=
s
+
len
/
2
;
for
(
int
i
=
threadIdx
.
x
;
;
i
+=
blockDim
.
x
)
//simulates other blocks in case src.size > blockDim.x*2
{
//calculates which 2 indexes will be compared and swap
int
part
=
i
/
(
len
/
2
);
int
s
=
part
*
len
+
(
i
&
((
len
/
2
)
-
1
));
int
e
=
s
+
len
/
2
;
if
(
e
>=
src
.
getSize
())
break
;
//calculate the direction of swapping
int
monotonicSeqIdx
=
i
/
(
monotonicSeqLen
/
2
);
bool
ascending
=
(
monotonicSeqIdx
&
1
)
!=
0
;
if
((
monotonicSeqIdx
+
1
)
*
monotonicSeqLen
>=
src
.
getSize
())
//special case for parts with no "partner"
ascending
=
true
;
if
(
e
<
src
.
getSize
())
//not touching virtual padding
cmpSwap
(
src
[
s
],
src
[
e
],
ascending
,
Cmp
);
}
__syncthreads
();
}
}
...
...
@@ -292,9 +300,9 @@ __global__ void bitoniSort1stStep(TNL::Containers::ArrayView<Value, TNL::Devices
int
myBlockEnd
=
TNL
::
min
(
arr
.
getSize
(),
myBlockStart
+
(
2
*
blockDim
.
x
));
if
(
blockIdx
.
x
%
2
||
blockIdx
.
x
+
1
==
gridDim
.
x
)
bitonicSort_Block
(
arr
.
getView
(
myBlockStart
,
myBlockEnd
),
arr
.
getView
(
myBlockStart
,
myBlockEnd
),
Cmp
);
bitonicSort_Block
(
arr
.
getView
(
myBlockStart
,
myBlockEnd
),
Cmp
);
else
bitonicSort_Block
(
arr
.
getView
(
myBlockStart
,
myBlockEnd
),
arr
.
getView
(
myBlockStart
,
myBlockEnd
),
bitonicSort_Block
(
arr
.
getView
(
myBlockStart
,
myBlockEnd
),
[
&
]
__cuda_callable__
(
const
Value
&
a
,
const
Value
&
b
)
{
return
Cmp
(
b
,
a
);
});
}
...
...
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Save comment
Cancel
Please
register
or
sign in
to comment