Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
PaddlePaddle
PaddleDetection
提交
36444461
P
PaddleDetection
项目概览
PaddlePaddle
/
PaddleDetection
1 年多 前同步成功
通知
696
Star
11112
Fork
2696
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
184
列表
看板
标记
里程碑
合并请求
40
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
PaddleDetection
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
184
Issue
184
列表
看板
标记
里程碑
合并请求
40
合并请求
40
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
未验证
提交
36444461
编写于
12月 05, 2017
作者:
C
chengduo
提交者:
GitHub
12月 05, 2017
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #6229 from chengduoZH/profiling/updata_elementwise_op
[Profiling] Update elementwise op
上级
1b6804fb
37671ac0
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
190 addition
and
1 deletion
+190
-1
paddle/operators/elementwise_add_op.h
paddle/operators/elementwise_add_op.h
+38
-1
paddle/operators/elementwise_op_function.h
paddle/operators/elementwise_op_function.h
+152
-0
未找到文件。
paddle/operators/elementwise_add_op.h
浏览文件 @
36444461
...
...
@@ -19,11 +19,48 @@
namespace
paddle
{
namespace
operators
{
template
<
typename
T
>
struct
AddFunctor
{
HOSTDEVICE
T
operator
()(
T
a
,
T
b
)
const
{
return
a
+
b
;
}
};
template
<
typename
Place
,
typename
T
>
class
ElementwiseAddKernel
:
public
framework
::
OpKernel
<
T
>
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
ctx
)
const
override
{
ElementwiseCompute
<
EigenAddFunctor
,
Place
,
T
>
(
ctx
);
using
Tensor
=
framework
::
Tensor
;
auto
*
x
=
ctx
.
Input
<
Tensor
>
(
"X"
);
auto
*
y
=
ctx
.
Input
<
Tensor
>
(
"Y"
);
auto
*
z
=
ctx
.
Output
<
Tensor
>
(
"Out"
);
z
->
mutable_data
<
T
>
(
ctx
.
GetPlace
());
TransformFunctor
<
AddFunctor
<
T
>
,
T
,
Place
>
functor
(
x
,
y
,
z
,
ctx
.
device_context
(),
AddFunctor
<
T
>
());
auto
x_dims
=
x
->
dims
();
auto
y_dims
=
y
->
dims
();
PADDLE_ENFORCE_GE
(
x_dims
.
size
(),
y_dims
.
size
(),
"Rank of first input must >= rank of second input."
);
if
(
x_dims
==
y_dims
)
{
functor
.
Run
();
return
;
}
int
axis
=
ctx
.
Attr
<
int
>
(
"axis"
);
axis
=
(
axis
==
-
1
?
x_dims
.
size
()
-
y_dims
.
size
()
:
axis
);
PADDLE_ENFORCE
(
axis
>=
0
&&
axis
<
x_dims
.
size
(),
"Axis should be in range [0, x_dims)"
);
int
pre
,
n
,
post
;
get_mid_dims
(
x_dims
,
y_dims
,
axis
,
pre
,
n
,
post
);
if
(
post
==
1
)
{
functor
.
RunRowWise
(
n
,
pre
);
return
;
}
else
{
functor
.
RunMidWise
(
n
,
pre
,
post
);
return
;
}
}
};
...
...
paddle/operators/elementwise_op_function.h
浏览文件 @
36444461
...
...
@@ -16,6 +16,11 @@
#include "paddle/framework/eigen.h"
#include "paddle/framework/op_registry.h"
#include "paddle/framework/operator.h"
#include "paddle/platform/transform.h"
#ifdef __NVCC__
#include <thrust/iterator/iterator_adaptor.h>
#endif
#include "paddle/operators/math/math_function.h"
...
...
@@ -54,6 +59,153 @@ inline void get_mid_dims(const framework::DDim& x_dims,
}
}
template
<
typename
T
,
typename
Place
>
class
RowwiseTransformIterator
;
template
<
typename
T
,
typename
Place
>
class
MidWiseTransformIterator
;
template
<
typename
T
>
class
RowwiseTransformIterator
<
T
,
platform
::
CPUPlace
>
{
public:
RowwiseTransformIterator
(
const
T
*
ptr
,
int
n
)
:
ptr_
(
ptr
),
i_
(
0
),
n_
(
n
)
{}
RowwiseTransformIterator
<
T
,
platform
::
CPUPlace
>&
operator
++
()
{
++
i_
;
i_
%=
n_
;
return
*
this
;
}
bool
operator
==
(
const
RowwiseTransformIterator
<
T
,
platform
::
CPUPlace
>&
rhs
)
const
{
return
(
ptr_
+
i_
)
==
&
(
*
rhs
);
}
bool
operator
!=
(
const
RowwiseTransformIterator
<
T
,
platform
::
CPUPlace
>&
rhs
)
const
{
return
(
ptr_
+
i_
)
!=
&
(
*
rhs
);
}
const
T
&
operator
*
()
{
return
ptr_
[
i_
];
}
private:
const
T
*
ptr_
;
int
i_
;
int64_t
n_
;
};
template
<
typename
T
>
class
MidWiseTransformIterator
<
T
,
platform
::
CPUPlace
>
{
public:
MidWiseTransformIterator
(
const
T
*
ptr
,
int
n
,
int
post
)
:
ptr_
(
ptr
),
i_
(
0
),
j_
(
0
),
n_
(
n
),
post_
(
post
)
{}
MidWiseTransformIterator
<
T
,
platform
::
CPUPlace
>&
operator
++
()
{
i_
=
(
++
j_
/
post_
)
%
n_
;
return
*
this
;
}
bool
operator
==
(
const
MidWiseTransformIterator
<
T
,
platform
::
CPUPlace
>&
rhs
)
const
{
return
(
ptr_
+
i_
)
==
&
(
*
rhs
);
}
bool
operator
!=
(
const
MidWiseTransformIterator
<
T
,
platform
::
CPUPlace
>&
rhs
)
const
{
return
(
ptr_
+
i_
)
!=
&
(
*
rhs
);
}
const
T
&
operator
*
()
{
return
ptr_
[
i_
];
}
private:
const
T
*
ptr_
;
int
i_
;
int64_t
j_
;
int64_t
n_
;
int
post_
;
};
#ifdef __NVCC__
template
<
typename
T
>
class
RowwiseTransformIterator
<
T
,
platform
::
GPUPlace
>
:
public
thrust
::
iterator_adaptor
<
RowwiseTransformIterator
<
T
,
platform
::
GPUPlace
>
,
const
T
*>
{
public:
typedef
thrust
::
iterator_adaptor
<
RowwiseTransformIterator
<
T
,
platform
::
GPUPlace
>
,
const
T
*>
super_t
;
HOSTDEVICE
RowwiseTransformIterator
(
const
T
*
x
,
int
n
)
:
super_t
(
x
),
begin_
(
x
),
n_
(
n
){};
friend
class
thrust
::
iterator_core_access
;
private:
unsigned
int
n_
;
const
T
*
begin_
;
HOSTDEVICE
typename
super_t
::
reference
dereference
()
const
{
return
*
(
begin_
+
(
this
->
base
()
-
begin_
)
%
n_
);
}
};
template
<
typename
T
>
class
MidWiseTransformIterator
<
T
,
platform
::
GPUPlace
>
:
public
thrust
::
iterator_adaptor
<
MidWiseTransformIterator
<
T
,
platform
::
GPUPlace
>
,
const
T
*>
{
public:
typedef
thrust
::
iterator_adaptor
<
MidWiseTransformIterator
<
T
,
platform
::
GPUPlace
>
,
const
T
*>
super_t
;
HOSTDEVICE
MidWiseTransformIterator
(
const
T
*
x
,
int
n
,
int
post
)
:
super_t
(
x
),
begin_
(
x
),
n_
(
n
),
post_
(
post
){};
friend
class
thrust
::
iterator_core_access
;
private:
unsigned
int
post_
;
unsigned
int
n_
;
const
T
*
begin_
;
HOSTDEVICE
typename
super_t
::
reference
dereference
()
const
{
return
*
(
begin_
+
(((
this
->
base
()
-
begin_
)
/
post_
)
%
n_
));
}
};
#endif
template
<
typename
Functor
,
typename
T
,
typename
Place
>
class
TransformFunctor
{
public:
TransformFunctor
(
const
framework
::
Tensor
*
x
,
const
framework
::
Tensor
*
y
,
framework
::
Tensor
*
z
,
const
platform
::
DeviceContext
&
ctx
,
Functor
func
)
:
x_
(
x
->
data
<
T
>
()),
y_
(
y
->
data
<
T
>
()),
z_
(
z
->
mutable_data
<
T
>
(
ctx
.
GetPlace
())),
nx_
(
x
->
numel
()),
ctx_
(
ctx
),
func_
(
func
)
{}
inline
void
Run
()
const
{
platform
::
Transform
<
Place
>
trans
;
trans
(
ctx_
,
x_
,
x_
+
nx_
,
y_
,
z_
,
func_
);
}
inline
void
RunRowWise
(
int
n
,
int
pre
)
const
{
platform
::
Transform
<
Place
>
trans
;
trans
(
ctx_
,
x_
,
x_
+
nx_
,
RowwiseTransformIterator
<
T
,
Place
>
(
y_
,
n
),
z_
,
func_
);
}
inline
void
RunMidWise
(
int
n
,
int
pre
,
int
post
)
const
{
platform
::
Transform
<
Place
>
trans
;
trans
(
ctx_
,
x_
,
x_
+
nx_
,
MidWiseTransformIterator
<
T
,
Place
>
(
y_
,
n
,
post
),
z_
,
func_
);
}
private:
const
T
*
x_
;
const
T
*
y_
;
T
*
z_
;
int64_t
nx_
;
const
platform
::
DeviceContext
&
ctx_
;
Functor
func_
;
};
#define EIGEN_FUNCTOR(name, eigen_op) \
struct Eigen##name##Functor { \
template <typename Place, typename T> \
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录