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
urhero
Paddle
提交
6be273cb
P
Paddle
项目概览
urhero
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
0
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
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
0
Issue
0
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
体验新版 GitCode,发现更多精彩内容 >>
提交
6be273cb
编写于
8月 24, 2018
作者:
T
tensor-tang
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
add seq mode lstm
上级
36363292
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
45 addition
and
7 deletion
+45
-7
paddle/fluid/operators/fusion_lstm_op.cc
paddle/fluid/operators/fusion_lstm_op.cc
+45
-7
未找到文件。
paddle/fluid/operators/fusion_lstm_op.cc
浏览文件 @
6be273cb
...
...
@@ -15,10 +15,14 @@ limitations under the License. */
#include "paddle/fluid/operators/fusion_lstm_op.h"
#include <string>
#include "paddle/fluid/operators/math/blas.h"
#include "paddle/fluid/operators/math/cpu_vec.h"
#include "paddle/fluid/operators/math/detail/activation_functions.h"
#include "paddle/fluid/operators/math/fc_compute.h"
#include "paddle/fluid/operators/math/lstm_compute.h"
#include "paddle/fluid/operators/math/sequence2batch.h"
#include "paddle/fluid/platform/cpu_info.h"
DEFINE_bool
(
seq_mode
,
true
,
"Use sequence mode"
);
namespace
paddle
{
namespace
operators
{
...
...
@@ -98,7 +102,12 @@ void FusionLSTMOp::InferShape(framework::InferShapeContext* ctx) const {
ctx
->
ShareLoD
(
"X"
,
"Hidden"
);
ctx
->
ShareLoD
(
"X"
,
"Cell"
);
int
xx_width
=
x_dims
[
1
]
>
wx_dims
[
1
]
?
wx_dims
[
1
]
:
x_dims
[
1
];
int
xx_width
;
if
(
FLAGS_seq_mode
)
{
xx_width
=
wx_dims
[
1
];
}
else
{
xx_width
=
x_dims
[
1
]
>
wx_dims
[
1
]
?
wx_dims
[
1
]
:
x_dims
[
1
];
}
ctx
->
SetOutputDim
(
"XX"
,
{
x_dims
[
0
],
xx_width
});
ctx
->
ShareLoD
(
"X"
,
"XX"
);
}
...
...
@@ -205,10 +214,34 @@ inline void ReorderInitState(const DeviceContext& ctx,
row_shuffle
(
ctx
,
src
,
index_lod
,
dst
,
indexed_src
);
}
template
<
typename
DeviceContext
,
typename
T
>
template
<
typename
T
>
class
FuisonLSTMKernel
:
public
framework
::
OpKernel
<
T
>
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
ctx
)
const
override
{
void
SeqCompute
(
const
framework
::
ExecutionContext
&
ctx
)
const
{
using
DeviceContext
=
paddle
::
platform
::
CPUDeviceContext
;
auto
*
x
=
ctx
.
Input
<
LoDTensor
>
(
"X"
);
auto
*
wx
=
ctx
.
Input
<
Tensor
>
(
"WeightX"
);
auto
*
wh
=
ctx
.
Input
<
Tensor
>
(
"WeightH"
);
auto
*
bias
=
ctx
.
Input
<
Tensor
>
(
"Bias"
);
auto
*
xx
=
ctx
.
Output
<
LoDTensor
>
(
"XX"
);
auto
x_dims
=
x
->
dims
();
// T x M
auto
wh_dims
=
wh
->
dims
();
// D x 4D
const
int
M
=
x_dims
[
1
];
// x frame size
const
int
D4
=
wh_dims
[
1
];
const
T
*
x_data
=
x
->
data
<
T
>
();
const
T
*
wx_data
=
wx
->
data
<
T
>
();
T
*
xx_data
=
xx
->
mutable_data
<
T
>
(
ctx
.
GetPlace
());
auto
blas
=
math
::
GetBlas
<
DeviceContext
,
T
>
(
ctx
);
math
::
FCCompute
<
DeviceContext
,
T
>
(
blas
,
x_dims
[
0
],
D4
,
M
,
x_data
,
wx_data
,
xx_data
,
bias
->
data
<
T
>
());
}
void
BatchCompute
(
const
framework
::
ExecutionContext
&
ctx
)
const
{
using
DeviceContext
=
platform
::
CPUDeviceContext
;
auto
*
x
=
ctx
.
Input
<
LoDTensor
>
(
"X"
);
auto
*
wx
=
ctx
.
Input
<
Tensor
>
(
"WeightX"
);
auto
*
wh
=
ctx
.
Input
<
Tensor
>
(
"WeightH"
);
...
...
@@ -339,6 +372,13 @@ class FuisonLSTMKernel : public framework::OpKernel<T> {
// restore the output cell state in LoDTensor from the batch cell
to_seq
(
dev_ctx
,
batch_cell
,
cell_out
);
}
void
Compute
(
const
framework
::
ExecutionContext
&
ctx
)
const
override
{
if
(
FLAGS_seq_mode
)
{
SeqCompute
(
ctx
);
}
else
{
BatchCompute
(
ctx
);
}
}
};
}
// namespace operators
...
...
@@ -348,7 +388,5 @@ namespace ops = paddle::operators;
REGISTER_OPERATOR
(
fusion_lstm
,
ops
::
FusionLSTMOp
,
ops
::
FusionLSTMOpMaker
,
paddle
::
framework
::
DefaultGradOpDescMaker
<
true
>
);
REGISTER_OP_CPU_KERNEL
(
fusion_lstm
,
ops
::
FuisonLSTMKernel
<
paddle
::
platform
::
CPUDeviceContext
,
float
>
,
ops
::
FuisonLSTMKernel
<
paddle
::
platform
::
CPUDeviceContext
,
double
>
);
REGISTER_OP_CPU_KERNEL
(
fusion_lstm
,
ops
::
FuisonLSTMKernel
<
float
>
,
ops
::
FuisonLSTMKernel
<
double
>
);
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