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
PaddleDetection
提交
a7b0d5bd
P
PaddleDetection
项目概览
PaddlePaddle
/
PaddleDetection
大约 1 年 前同步成功
通知
694
Star
11112
Fork
2696
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
184
列表
看板
标记
里程碑
合并请求
40
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
PaddleDetection
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
184
Issue
184
列表
看板
标记
里程碑
合并请求
40
合并请求
40
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
a7b0d5bd
编写于
3月 24, 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
Clean code
上级
e3144393
变更
2
显示空白变更内容
内联
并排
Showing
2 changed file
with
10 addition
and
13 deletion
+10
-13
paddle/fluid/framework/parallel_executor.cc
paddle/fluid/framework/parallel_executor.cc
+8
-11
paddle/fluid/framework/parallel_executor.h
paddle/fluid/framework/parallel_executor.h
+2
-2
未找到文件。
paddle/fluid/framework/parallel_executor.cc
浏览文件 @
a7b0d5bd
...
@@ -27,15 +27,16 @@ namespace framework {
...
@@ -27,15 +27,16 @@ namespace framework {
class
ParallelExecutorPrivate
{
class
ParallelExecutorPrivate
{
public:
public:
explicit
ParallelExecutorPrivate
(
const
std
::
vector
<
platform
::
Place
>
&
places
)
explicit
ParallelExecutorPrivate
(
const
std
::
vector
<
platform
::
Place
>
&
places
)
:
places_
(
places
)
,
fetch_dev_ctxs_
(
places
)
{}
:
places_
(
places
)
{}
std
::
vector
<
platform
::
Place
>
places_
;
std
::
vector
<
platform
::
Place
>
places_
;
platform
::
DeviceContextPool
fetch_dev_ctxs_
;
std
::
vector
<
Scope
*>
local_scopes_
;
std
::
vector
<
Scope
*>
local_scopes_
;
Scope
*
global_scope_
;
Scope
*
global_scope_
;
std
::
unique_ptr
<
details
::
SSAGraphExecutor
>
executor_
;
#ifdef PADDLE_WITH_CUDA
std
::
unique_ptr
<
platform
::
NCCLContextMap
>
nccl_ctxs_
;
std
::
unique_ptr
<
platform
::
NCCLContextMap
>
nccl_ctxs_
;
std
::
unique_ptr
<
details
::
SSAGraphExecutor
>
executor_
;
#endif
};
};
ParallelExecutor
::
ParallelExecutor
(
ParallelExecutor
::
ParallelExecutor
(
...
@@ -54,8 +55,10 @@ ParallelExecutor::ParallelExecutor(
...
@@ -54,8 +55,10 @@ ParallelExecutor::ParallelExecutor(
member_
->
local_scopes_
.
push_back
(
&
scope
->
NewScope
());
member_
->
local_scopes_
.
push_back
(
&
scope
->
NewScope
());
}
}
// Bcast Parameters to all GPUs
// Bcast Parameters to all GPUs
BuildNCCLCommunicator
();
#ifdef PADDLE_WITH_CUDA
member_
->
nccl_ctxs_
.
reset
(
new
platform
::
NCCLContextMap
(
member_
->
places_
));
#endif
if
(
platform
::
is_gpu_place
(
places
[
0
])
&&
if
(
platform
::
is_gpu_place
(
places
[
0
])
&&
member_
->
local_scopes_
.
size
()
!=
1
)
{
// Is CUDA
member_
->
local_scopes_
.
size
()
!=
1
)
{
// Is CUDA
BCastParamsToGPUs
(
startup_program
);
BCastParamsToGPUs
(
startup_program
);
...
@@ -123,12 +126,6 @@ void ParallelExecutor::BCastParamsToGPUs(
...
@@ -123,12 +126,6 @@ void ParallelExecutor::BCastParamsToGPUs(
#endif
#endif
}
}
void
ParallelExecutor
::
BuildNCCLCommunicator
()
const
{
#ifdef PADDLE_WITH_CUDA
member_
->
nccl_ctxs_
.
reset
(
new
platform
::
NCCLContextMap
(
member_
->
places_
));
#endif
}
void
ParallelExecutor
::
Run
(
const
std
::
vector
<
std
::
string
>
&
fetch_tensors
,
void
ParallelExecutor
::
Run
(
const
std
::
vector
<
std
::
string
>
&
fetch_tensors
,
const
std
::
string
&
fetched_var_name
)
{
const
std
::
string
&
fetched_var_name
)
{
auto
fetch_data
=
member_
->
executor_
->
Run
(
fetch_tensors
);
auto
fetch_data
=
member_
->
executor_
->
Run
(
fetch_tensors
);
...
...
paddle/fluid/framework/parallel_executor.h
浏览文件 @
a7b0d5bd
...
@@ -31,6 +31,8 @@ namespace framework {
...
@@ -31,6 +31,8 @@ namespace framework {
class
ParallelExecutorPrivate
;
class
ParallelExecutorPrivate
;
class
ParallelExecutor
{
class
ParallelExecutor
{
DISABLE_COPY_AND_ASSIGN
(
ParallelExecutor
);
public:
public:
explicit
ParallelExecutor
(
size_t
num_threads
,
explicit
ParallelExecutor
(
size_t
num_threads
,
const
std
::
vector
<
platform
::
Place
>&
places
,
const
std
::
vector
<
platform
::
Place
>&
places
,
...
@@ -46,8 +48,6 @@ class ParallelExecutor {
...
@@ -46,8 +48,6 @@ class ParallelExecutor {
ParallelExecutorPrivate
*
member_
;
ParallelExecutorPrivate
*
member_
;
void
BCastParamsToGPUs
(
const
ProgramDesc
&
startup_program
)
const
;
void
BCastParamsToGPUs
(
const
ProgramDesc
&
startup_program
)
const
;
void
BuildNCCLCommunicator
()
const
;
};
};
}
// namespace framework
}
// namespace framework
...
...
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