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
OPTHREE
Paddle
提交
9824e8f3
P
Paddle
项目概览
OPTHREE
/
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
体验新版 GitCode,发现更多精彩内容 >>
提交
9824e8f3
编写于
3月 19, 2018
作者:
Y
Yu Yang
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Scale loss op use event
上级
071043c3
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
18 addition
and
6 deletion
+18
-6
paddle/fluid/framework/parallel_executor.cc
paddle/fluid/framework/parallel_executor.cc
+18
-6
未找到文件。
paddle/fluid/framework/parallel_executor.cc
浏览文件 @
9824e8f3
...
@@ -124,12 +124,17 @@ struct ScaleLossGradOpHandle : public OpHandle {
...
@@ -124,12 +124,17 @@ struct ScaleLossGradOpHandle : public OpHandle {
float
coeff_
;
float
coeff_
;
Scope
*
scope_
;
Scope
*
scope_
;
platform
::
Place
place_
;
platform
::
Place
place_
;
cudaEvent_t
ev_
;
explicit
ScaleLossGradOpHandle
(
size_t
num_dev
,
Scope
*
scope
,
explicit
ScaleLossGradOpHandle
(
size_t
num_dev
,
Scope
*
scope
,
platform
::
Place
place
)
platform
::
Place
place
)
:
coeff_
(
static_cast
<
float
>
(
1.0
/
num_dev
)),
:
coeff_
(
static_cast
<
float
>
(
1.0
/
num_dev
)),
scope_
(
scope
),
scope_
(
scope
),
place_
(
place
)
{}
place_
(
place
)
{
PADDLE_ENFORCE
(
cudaEventCreateWithFlags
(
&
ev_
,
cudaEventDisableTiming
));
}
~
ScaleLossGradOpHandle
()
{
PADDLE_ENFORCE
(
cudaEventDestroy
(
ev_
));
}
void
Run
()
override
{
void
Run
()
override
{
std
::
string
var_name
=
static_cast
<
VarHandle
*>
(
this
->
outputs_
[
0
])
->
name_
;
std
::
string
var_name
=
static_cast
<
VarHandle
*>
(
this
->
outputs_
[
0
])
->
name_
;
...
@@ -141,16 +146,23 @@ struct ScaleLossGradOpHandle : public OpHandle {
...
@@ -141,16 +146,23 @@ struct ScaleLossGradOpHandle : public OpHandle {
if
(
platform
::
is_cpu_place
(
place_
))
{
if
(
platform
::
is_cpu_place
(
place_
))
{
*
tmp
=
coeff_
;
*
tmp
=
coeff_
;
}
else
{
}
else
{
memory
::
Copy
(
auto
stream
=
boost
::
get
<
platform
::
CUDAPlace
>
(
place_
),
tmp
,
platform
::
CPUPlace
(),
&
coeff_
,
sizeof
(
float
),
static_cast
<
platform
::
CUDADeviceContext
*>
(
this
->
dev_ctx_
[
place_
])
static_cast
<
platform
::
CUDADeviceContext
*>
(
this
->
dev_ctx_
[
place_
])
->
stream
());
->
stream
();
memory
::
Copy
(
boost
::
get
<
platform
::
CUDAPlace
>
(
place_
),
tmp
,
platform
::
CPUPlace
(),
&
coeff_
,
sizeof
(
float
),
stream
);
PADDLE_ENFORCE
(
cudaEventRecord
(
ev_
,
stream
));
}
}
}
}
void
Wait
(
platform
::
DeviceContext
*
waited_dev
)
override
{
void
Wait
(
platform
::
DeviceContext
*
waited_dev
)
override
{
this
->
dev_ctx_
.
at
(
place_
)
->
Wait
();
if
(
platform
::
is_cpu_place
(
waited_dev
->
GetPlace
()))
{
this
->
dev_ctx_
.
at
(
place_
)
->
Wait
();
}
else
{
auto
stream
=
static_cast
<
platform
::
CUDADeviceContext
*>
(
waited_dev
)
->
stream
();
PADDLE_ENFORCE
(
cudaStreamWaitEvent
(
stream
,
ev_
,
0
));
}
}
}
};
};
...
...
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