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
2ecc5622
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
2ecc5622
编写于
6月 28, 2018
作者:
X
Xin Pan
提交者:
whs
6月 28, 2018
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
small AverageOptimizer enhance. (#11761)
* small AverageOptimizer enhance. * clean * clean
上级
19e877ff
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
14 addition
and
13 deletion
+14
-13
paddle/fluid/operators/average_accumulates_op.cc
paddle/fluid/operators/average_accumulates_op.cc
+11
-11
paddle/fluid/operators/average_accumulates_op.h
paddle/fluid/operators/average_accumulates_op.h
+3
-2
未找到文件。
paddle/fluid/operators/average_accumulates_op.cc
浏览文件 @
2ecc5622
...
...
@@ -19,28 +19,28 @@ namespace operators {
template
<
>
void
GetAccumulators
<
paddle
::
platform
::
CPUDeviceContext
>
(
const
framework
::
ExecutionContext
&
ctx
,
int64_t
*
num_updates
_
,
int64_t
*
num_accumulates
_
,
int64_t
*
old_num_accumulates_
)
{
const
framework
::
ExecutionContext
&
ctx
,
int64_t
*
num_updates
,
int64_t
*
num_accumulates
,
int64_t
*
old_num_accumulates
)
{
auto
*
in_old_num_accumulates
=
ctx
.
Input
<
Tensor
>
(
"in_old_num_accumulates"
);
auto
*
in_num_accumulates
=
ctx
.
Input
<
Tensor
>
(
"in_num_accumulates"
);
auto
*
in_num_updates
=
ctx
.
Input
<
Tensor
>
(
"in_num_updates"
);
*
old_num_accumulates
_
=
in_old_num_accumulates
->
data
<
int64_t
>
()[
0
];
*
num_accumulates
_
=
in_num_accumulates
->
data
<
int64_t
>
()[
0
];
*
num_updates
_
=
in_num_updates
->
data
<
int64_t
>
()[
0
];
*
old_num_accumulates
=
in_old_num_accumulates
->
data
<
int64_t
>
()[
0
];
*
num_accumulates
=
in_num_accumulates
->
data
<
int64_t
>
()[
0
];
*
num_updates
=
in_num_updates
->
data
<
int64_t
>
()[
0
];
}
template
<
>
void
SetAccumulators
<
paddle
::
platform
::
CPUDeviceContext
>
(
const
framework
::
ExecutionContext
&
ctx
,
int64_t
num_updates
_
,
int64_t
num_accumulates
_
,
int64_t
old_num_accumulates_
)
{
const
framework
::
ExecutionContext
&
ctx
,
int64_t
num_updates
,
int64_t
num_accumulates
,
int64_t
old_num_accumulates
)
{
auto
*
out_old_num_accumulates
=
ctx
.
Output
<
Tensor
>
(
"out_old_num_accumulates"
);
auto
*
out_num_accumulates
=
ctx
.
Output
<
Tensor
>
(
"out_num_accumulates"
);
auto
*
out_num_updates
=
ctx
.
Output
<
Tensor
>
(
"out_num_updates"
);
out_old_num_accumulates
->
data
<
int64_t
>
()[
0
]
=
old_num_accumulates
_
;
out_num_accumulates
->
data
<
int64_t
>
()[
0
]
=
num_accumulates
_
;
out_num_updates
->
data
<
int64_t
>
()[
0
]
=
num_updates
_
;
out_old_num_accumulates
->
data
<
int64_t
>
()[
0
]
=
old_num_accumulates
;
out_num_accumulates
->
data
<
int64_t
>
()[
0
]
=
num_accumulates
;
out_num_updates
->
data
<
int64_t
>
()[
0
]
=
num_updates
;
}
class
AverageAccumulatesOp
:
public
framework
::
OperatorWithKernel
{
...
...
@@ -177,7 +177,7 @@ class AverageAccumulatesOpMaker : public framework::OpProtoAndCheckerMaker {
AddComment
(
R"DOC(
AverageAccumulates Operator.
Accumulate the sum of parameter w
ht
in sliding window. The size of sliding window is
Accumulate the sum of parameter w
ith
in sliding window. The size of sliding window is
determined by 'average_window', 'max_average_window' and 'min_average_window'.
Memory was shared by Input(in_sum_1) and Output(out_sum_1) which acts as an accumulator 'sum_1'.
'sum_2', 'sum_3', 'num_accumulates', 'old_num_accumulates' and 'num_updates' were the same as 'sum_1'.
...
...
paddle/fluid/operators/average_accumulates_op.h
浏览文件 @
2ecc5622
...
...
@@ -54,8 +54,9 @@ class AverageAccumulatesKernel : public framework::OpKernel<T> {
float
average_window
=
ctx
.
Attr
<
float
>
(
"average_window"
);
int64_t
max_average_window
=
ctx
.
Attr
<
int64_t
>
(
"max_average_window"
);
int64_t
min_average_window
=
ctx
.
Attr
<
int64_t
>
(
"min_average_window"
);
min_average_window
=
std
::
min
<
int64_t
>
(
min_average_window
,
max_average_window
);
PADDLE_ENFORCE_LE
(
min_average_window
,
max_average_window
,
"min_average_window shouldn't be larger than "
"max_average_window"
);
// Get inputs
auto
*
param
=
ctx
.
Input
<
Tensor
>
(
"param"
);
...
...
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