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
机器未来
Paddle
提交
0385b0a1
P
Paddle
项目概览
机器未来
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
1
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1
列表
看板
标记
里程碑
合并请求
0
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
Paddle
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
1
Issue
1
列表
看板
标记
里程碑
合并请求
0
合并请求
0
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
提交
0385b0a1
编写于
10月 11, 2018
作者:
M
minqiyang
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Accelerate SequencePool Op on SUM mode
test=develop
上级
e1904ac2
变更
2
显示空白变更内容
内联
并排
Showing
2 changed file
with
20 addition
and
5 deletion
+20
-5
paddle/fluid/operators/math/CMakeLists.txt
paddle/fluid/operators/math/CMakeLists.txt
+2
-2
paddle/fluid/operators/math/sequence_pooling.cc
paddle/fluid/operators/math/sequence_pooling.cc
+18
-3
未找到文件。
paddle/fluid/operators/math/CMakeLists.txt
浏览文件 @
0385b0a1
paddle/fluid/operators/math/sequence_pooling.cc
浏览文件 @
0385b0a1
...
@@ -12,9 +12,11 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
...
@@ -12,9 +12,11 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
See the License for the specific language governing permissions and
limitations under the License. */
limitations under the License. */
#include "paddle/fluid/operators/math/sequence_pooling.h"
#include <string>
#include <string>
#include "paddle/fluid/operators/math/blas.h"
#include "paddle/fluid/operators/math/math_function.h"
#include "paddle/fluid/operators/math/math_function.h"
#include "paddle/fluid/operators/math/sequence_pooling.h"
namespace
paddle
{
namespace
paddle
{
namespace
operators
{
namespace
operators
{
...
@@ -180,6 +182,7 @@ class SequencePoolFunctor<platform::CPUDeviceContext, T> {
...
@@ -180,6 +182,7 @@ class SequencePoolFunctor<platform::CPUDeviceContext, T> {
}
}
auto
lod
=
input
.
lod
()[
0
];
auto
lod
=
input
.
lod
()[
0
];
auto
&
place
=
*
context
.
eigen_device
();
auto
&
place
=
*
context
.
eigen_device
();
auto
blas
=
math
::
GetBlas
<
platform
::
CPUDeviceContext
,
T
>
(
context
);
for
(
int
i
=
0
;
i
<
static_cast
<
int
>
(
lod
.
size
())
-
1
;
++
i
)
{
for
(
int
i
=
0
;
i
<
static_cast
<
int
>
(
lod
.
size
())
-
1
;
++
i
)
{
Tensor
in_t
=
Tensor
in_t
=
input
.
Slice
(
static_cast
<
int
>
(
lod
[
i
]),
static_cast
<
int
>
(
lod
[
i
+
1
]));
input
.
Slice
(
static_cast
<
int
>
(
lod
[
i
]),
static_cast
<
int
>
(
lod
[
i
+
1
]));
...
@@ -191,7 +194,14 @@ class SequencePoolFunctor<platform::CPUDeviceContext, T> {
...
@@ -191,7 +194,14 @@ class SequencePoolFunctor<platform::CPUDeviceContext, T> {
if
(
pooltype
==
"AVERAGE"
)
{
if
(
pooltype
==
"AVERAGE"
)
{
out_e
.
device
(
place
)
=
in_e
.
mean
(
Eigen
::
array
<
int
,
1
>
({{
0
}}));
out_e
.
device
(
place
)
=
in_e
.
mean
(
Eigen
::
array
<
int
,
1
>
({{
0
}}));
}
else
if
(
pooltype
==
"SUM"
)
{
}
else
if
(
pooltype
==
"SUM"
)
{
out_e
.
device
(
place
)
=
in_e
.
sum
(
Eigen
::
array
<
int
,
1
>
({{
0
}}));
if
(
h
>
0
)
{
const
T
*
in_data
=
in_t
.
data
<
T
>
();
T
*
out_data
=
out_t
.
mutable_data
<
T
>
(
context
.
GetPlace
());
blas
.
VCOPY
(
w
,
in_data
,
out_data
);
for
(
int64_t
r
=
1
;
r
!=
h
;
++
r
)
{
blas
.
AXPY
(
w
,
1.
,
in_data
+
r
*
w
,
out_data
);
}
}
}
else
if
(
pooltype
==
"SQRT"
)
{
}
else
if
(
pooltype
==
"SQRT"
)
{
out_e
.
device
(
place
)
=
in_e
.
sum
(
Eigen
::
array
<
int
,
1
>
({{
0
}}))
/
out_e
.
device
(
place
)
=
in_e
.
sum
(
Eigen
::
array
<
int
,
1
>
({{
0
}}))
/
std
::
sqrt
(
static_cast
<
T
>
(
h
));
std
::
sqrt
(
static_cast
<
T
>
(
h
));
...
@@ -223,6 +233,7 @@ class SequencePoolGradFunctor<platform::CPUDeviceContext, T> {
...
@@ -223,6 +233,7 @@ class SequencePoolGradFunctor<platform::CPUDeviceContext, T> {
}
}
auto
lod
=
in_grad
->
lod
()[
0
];
auto
lod
=
in_grad
->
lod
()[
0
];
auto
&
place
=
*
context
.
eigen_device
();
auto
&
place
=
*
context
.
eigen_device
();
auto
blas
=
math
::
GetBlas
<
platform
::
CPUDeviceContext
,
T
>
(
context
);
for
(
int
i
=
0
;
i
<
static_cast
<
int
>
(
lod
.
size
())
-
1
;
++
i
)
{
for
(
int
i
=
0
;
i
<
static_cast
<
int
>
(
lod
.
size
())
-
1
;
++
i
)
{
auto
in_g_t
=
in_grad
->
Slice
(
static_cast
<
int
>
(
lod
[
i
]),
auto
in_g_t
=
in_grad
->
Slice
(
static_cast
<
int
>
(
lod
[
i
]),
static_cast
<
int
>
(
lod
[
i
+
1
]));
static_cast
<
int
>
(
lod
[
i
+
1
]));
...
@@ -237,7 +248,11 @@ class SequencePoolGradFunctor<platform::CPUDeviceContext, T> {
...
@@ -237,7 +248,11 @@ class SequencePoolGradFunctor<platform::CPUDeviceContext, T> {
if
(
pooltype
==
"AVERAGE"
)
{
if
(
pooltype
==
"AVERAGE"
)
{
in_g_e
.
device
(
place
)
=
(
out_g_e
/
static_cast
<
T
>
(
h
)).
broadcast
(
bcast
);
in_g_e
.
device
(
place
)
=
(
out_g_e
/
static_cast
<
T
>
(
h
)).
broadcast
(
bcast
);
}
else
if
(
pooltype
==
"SUM"
)
{
}
else
if
(
pooltype
==
"SUM"
)
{
in_g_e
.
device
(
place
)
=
(
out_g_e
).
broadcast
(
bcast
);
const
T
*
out_g_data
=
out_g_t
.
data
<
T
>
();
T
*
in_g_data
=
in_g_t
.
mutable_data
<
T
>
(
context
.
GetPlace
());
for
(
int
r
=
0
;
r
!=
h
;
++
r
)
{
blas
.
VCOPY
(
w
,
out_g_data
,
in_g_data
+
r
*
w
);
}
}
else
if
(
pooltype
==
"SQRT"
)
{
}
else
if
(
pooltype
==
"SQRT"
)
{
in_g_e
.
device
(
place
)
=
in_g_e
.
device
(
place
)
=
(
out_g_e
/
std
::
sqrt
(
static_cast
<
T
>
(
h
))).
broadcast
(
bcast
);
(
out_g_e
/
std
::
sqrt
(
static_cast
<
T
>
(
h
))).
broadcast
(
bcast
);
...
...
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