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
Paddle
提交
0385b0a1
P
Paddle
项目概览
PaddlePaddle
/
Paddle
1 年多 前同步成功
通知
2302
Star
20931
Fork
5422
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1423
列表
看板
标记
里程碑
合并请求
543
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
1,423
Issue
1,423
列表
看板
标记
里程碑
合并请求
543
合并请求
543
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
6年前
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
隐藏空白更改
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
...
...
@@ -3,8 +3,8 @@ add_subdirectory(detail)
endif
(
NOT WIN32
)
function
(
math_library TARGET
)
# math_library is a function to create math library.
# The interface is the same as cc_library.
# math_library is a function to create math library.
# The interface is the same as cc_library.
# But it handle split GPU/CPU code and link some common library.
set
(
cc_srcs
)
set
(
cu_srcs
)
...
...
This diff is collapsed.
Click to expand it.
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.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/fluid/operators/math/sequence_pooling.h"
#include <string>
#include "paddle/fluid/operators/math/blas.h"
#include "paddle/fluid/operators/math/math_function.h"
#include "paddle/fluid/operators/math/sequence_pooling.h"
namespace
paddle
{
namespace
operators
{
...
...
@@ -180,6 +182,7 @@ class SequencePoolFunctor<platform::CPUDeviceContext, T> {
}
auto
lod
=
input
.
lod
()[
0
];
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
)
{
Tensor
in_t
=
input
.
Slice
(
static_cast
<
int
>
(
lod
[
i
]),
static_cast
<
int
>
(
lod
[
i
+
1
]));
...
...
@@ -191,7 +194,14 @@ class SequencePoolFunctor<platform::CPUDeviceContext, T> {
if
(
pooltype
==
"AVERAGE"
)
{
out_e
.
device
(
place
)
=
in_e
.
mean
(
Eigen
::
array
<
int
,
1
>
({{
0
}}));
}
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"
)
{
out_e
.
device
(
place
)
=
in_e
.
sum
(
Eigen
::
array
<
int
,
1
>
({{
0
}}))
/
std
::
sqrt
(
static_cast
<
T
>
(
h
));
...
...
@@ -223,6 +233,7 @@ class SequencePoolGradFunctor<platform::CPUDeviceContext, T> {
}
auto
lod
=
in_grad
->
lod
()[
0
];
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
)
{
auto
in_g_t
=
in_grad
->
Slice
(
static_cast
<
int
>
(
lod
[
i
]),
static_cast
<
int
>
(
lod
[
i
+
1
]));
...
...
@@ -237,7 +248,11 @@ class SequencePoolGradFunctor<platform::CPUDeviceContext, T> {
if
(
pooltype
==
"AVERAGE"
)
{
in_g_e
.
device
(
place
)
=
(
out_g_e
/
static_cast
<
T
>
(
h
)).
broadcast
(
bcast
);
}
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"
)
{
in_g_e
.
device
(
place
)
=
(
out_g_e
/
std
::
sqrt
(
static_cast
<
T
>
(
h
))).
broadcast
(
bcast
);
...
...
This diff is collapsed.
Click to expand it.
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
反馈
建议
客服
返回
顶部