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
efd5a849
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
体验新版 GitCode,发现更多精彩内容 >>
提交
efd5a849
编写于
7月 19, 2018
作者:
Y
Yancey1989
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
update executor interface
上级
800702cc
变更
6
隐藏空白更改
内联
并排
Showing
6 changed file
with
39 addition
and
13 deletion
+39
-13
paddle/fluid/framework/executor.cc
paddle/fluid/framework/executor.cc
+2
-2
paddle/fluid/framework/executor.h
paddle/fluid/framework/executor.h
+1
-1
paddle/fluid/operators/distributed/grpc_client.cc
paddle/fluid/operators/distributed/grpc_client.cc
+8
-4
paddle/fluid/operators/distributed/grpc_client.h
paddle/fluid/operators/distributed/grpc_client.h
+5
-1
paddle/fluid/pybind/pybind.cc
paddle/fluid/pybind/pybind.cc
+1
-3
python/paddle/fluid/executor.py
python/paddle/fluid/executor.py
+22
-2
未找到文件。
paddle/fluid/framework/executor.cc
浏览文件 @
efd5a849
...
...
@@ -45,13 +45,13 @@ ExecutorPrepareContext::~ExecutorPrepareContext() {
Executor
::
Executor
(
const
platform
::
Place
&
place
)
:
place_
(
place
)
{}
void
Executor
::
Close
()
{
#ifdef PADDLE_WITH_DISTRIBUTE
void
Executor
::
Complete
()
{
::
paddle
::
operators
::
distributed
::
RPCClient
::
GetInstance
<
::
paddle
::
operators
::
distributed
::
GRPCClient
>
()
->
SendComplete
();
}
#endif
}
void
InitializeVariable
(
Variable
*
var
,
proto
::
VarType
::
Type
var_type
)
{
if
(
var_type
==
proto
::
VarType
::
LOD_TENSOR
)
{
...
...
paddle/fluid/framework/executor.h
浏览文件 @
efd5a849
...
...
@@ -48,7 +48,7 @@ class Executor {
/*
* Sending signal to pserver to mark current trainer completed.
*/
void
C
omplet
e
();
void
C
los
e
();
#endif
...
...
paddle/fluid/operators/distributed/grpc_client.cc
浏览文件 @
efd5a849
...
...
@@ -36,11 +36,15 @@ void GRPCClient::InitEventLoop() {
}
void
GRPCClient
::
SendComplete
()
{
for
(
auto
&
it
:
channels_
)
{
VLOG
(
3
)
<<
"send complete message to "
<<
it
.
first
;
this
->
AsyncSendComplete
(
it
.
first
);
std
::
unique_lock
<
std
::
mutex
>
lk
(
completed_mutex_
);
if
(
!
completed_
)
{
for
(
auto
&
it
:
channels_
)
{
VLOG
(
3
)
<<
"send complete message to "
<<
it
.
first
;
this
->
AsyncSendComplete
(
it
.
first
);
}
PADDLE_ENFORCE
(
this
->
Wait
(),
"internal grpc error"
);
completed_
=
true
;
}
this
->
Wait
();
}
GRPCClient
::~
GRPCClient
()
{
...
...
paddle/fluid/operators/distributed/grpc_client.h
浏览文件 @
efd5a849
...
...
@@ -188,7 +188,7 @@ class CheckpointNotifyProcessor : public BaseProcessor {
class
GRPCClient
:
public
RPCClient
{
public:
GRPCClient
()
:
ok_
(
true
)
{}
GRPCClient
()
:
ok_
(
true
)
,
completed_
(
false
)
{}
virtual
~
GRPCClient
();
bool
AsyncSendVar
(
const
std
::
string
&
ep
,
const
platform
::
DeviceContext
&
ctx
,
...
...
@@ -247,6 +247,10 @@ class GRPCClient : public RPCClient {
// mutex for GetChannel thread safety
std
::
mutex
chan_mutex_
;
DISABLE_COPY_AND_ASSIGN
(
GRPCClient
);
// mutex for sending complete message only once
std
::
mutex
completed_mutex_
;
bool
completed_
;
};
}
// namespace distributed
...
...
paddle/fluid/pybind/pybind.cc
浏览文件 @
efd5a849
...
...
@@ -502,9 +502,7 @@ All parameter, weight, gradient are variables in Paddle.
py
::
class_
<
framework
::
Executor
>
(
m
,
"Executor"
)
.
def
(
py
::
init
<
const
platform
::
Place
&>
())
#ifdef PADDLE_WITH_DISTRIBUTE
.
def
(
"complete"
,
&
Executor
::
Complete
)
#endif
.
def
(
"close"
,
&
Executor
::
Close
)
.
def
(
"run"
,
[](
Executor
&
self
,
const
ProgramDesc
&
prog
,
Scope
*
scope
,
int
block_id
,
bool
create_local_scope
,
bool
create_vars
)
{
pybind11
::
gil_scoped_release
release
;
...
...
python/paddle/fluid/executor.py
浏览文件 @
efd5a849
...
...
@@ -247,6 +247,7 @@ class Executor(object):
p
.
set_place
(
place
)
self
.
executor
=
core
.
Executor
(
p
)
self
.
program_caches
=
dict
()
self
.
_closed
=
False
def
as_lodtensor
(
self
,
data
):
"""
...
...
@@ -348,8 +349,23 @@ class Executor(object):
]
return
outs
def
complete
(
self
):
self
.
executor
.
complete
()
def
close
(
self
):
"""
Close this executor.
You can no long use this executor after calling this method.
For the distributed training, this method would free the resource on PServers related to
the current Trainer.
Example:
>>> cpu = core.CPUPlace()
>>> exe = Executor(cpu)
>>> ...
>>> exe.close()
"""
if
not
self
.
_closed
:
self
.
executor
.
close
()
self
.
_closed
=
True
def
run
(
self
,
program
=
None
,
...
...
@@ -402,6 +418,10 @@ class Executor(object):
>>> feed={'X': x},
>>> fetch_list=[loss.name])
"""
if
self
.
_closed
:
raise
RuntimeError
(
"Attempted to use a closed Executor"
)
if
feed
is
None
:
feed
=
{}
if
not
isinstance
(
feed
,
dict
):
...
...
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