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
139ae08f
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
139ae08f
编写于
3月 20, 2018
作者:
T
typhoonzero
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
workable
上级
093e07d3
变更
1
显示空白变更内容
内联
并排
Showing
1 changed file
with
10 addition
and
16 deletion
+10
-16
paddle/fluid/operators/listen_and_serv_op.cc
paddle/fluid/operators/listen_and_serv_op.cc
+10
-16
未找到文件。
paddle/fluid/operators/listen_and_serv_op.cc
浏览文件 @
139ae08f
...
...
@@ -91,6 +91,9 @@ class ListenAndServOp : public framework::OperatorBase {
auto
*
block
=
Attr
<
framework
::
BlockDesc
*>
(
kOptimizeBlock
);
auto
*
program
=
block
->
Program
();
int
num_blocks
=
program
->
Size
();
PADDLE_ENFORCE_GE
(
num_blocks
,
2
,
"server program should have at least 2 blocks"
);
framework
::
Executor
executor
(
dev_place
);
// TODO(typhoonzero): change this to a while_op for every cluster-batch.
...
...
@@ -139,39 +142,30 @@ class ListenAndServOp : public framework::OperatorBase {
// should be global ops.
// NOTE: if is_gpu_place, CUDA kernels are laugched by multiple threads
// and this will still work.
double
ts
=
detail
::
GetTimestamp
();
std
::
vector
<
std
::
future
<
void
>>
fs
;
for
(
int
blkid
=
0
;
blkid
<
num_blocks
-
1
;
++
blkid
)
{
fs
.
push_back
(
framework
::
Async
([
&
]()
{
// block0 contains only listen_and_serv op, start run from block1.
for
(
int
blkid
=
1
;
blkid
<
num_blocks
-
1
;
++
blkid
)
{
fs
.
push_back
(
framework
::
Async
([
&
executor
,
&
program
,
&
recv_scope
,
blkid
]()
{
int
run_block
=
blkid
;
// thread local
try
{
VLOG
(
2
)
<<
"begin run in thread"
<<
blkid
;
executor
.
Run
(
*
program
,
&
recv_scope
,
blkid
,
executor
.
Run
(
*
program
,
&
recv_scope
,
run_block
,
false
/*create_local_scope*/
,
false
/*create_vars*/
);
VLOG
(
2
)
<<
"end run in thread"
;
}
catch
(
std
::
exception
&
e
)
{
LOG
(
ERROR
)
<<
"run sub program error "
<<
e
.
what
();
}
}));
}
VLOG
(
2
)
<<
"waiting opts..."
;
for
(
int
blkid
=
0
;
blkid
<
num_blocks
-
1
;
++
blkid
)
fs
[
blkid
].
wait
();
VLOG
(
2
)
<<
"waiting opts...OK"
;
for
(
int
i
=
0
;
i
<
num_blocks
-
2
;
++
i
)
fs
[
i
].
wait
();
// Run global block at final step
if
(
num_blocks
>
2
)
{
try
{
executor
.
Run
(
*
program
,
&
recv_scope
,
num_blocks
-
1
,
false
/*create_local_scope*/
,
false
/*create_vars*/
);
VLOG
(
2
)
<<
"run global OK , spent "
<<
detail
::
GetTimestamp
()
-
ts
;
}
catch
(
std
::
exception
&
e
)
{
LOG
(
ERROR
)
<<
"run sub program error "
<<
e
.
what
();
}
}
for
(
auto
&
n
:
recv_scope
.
LocalVarNames
())
{
VLOG
(
2
)
<<
"vars in scope: "
<<
n
;
}
for
(
auto
&
n
:
recv_scope
.
LocalVarNames
())
{
VLOG
(
2
)
<<
"vars in parent scope: "
<<
n
;
}
// Reset the received sparse variables, the sum operator would not
// sum the input sparse variables which rows is empty at the next
...
...
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