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
BaiXuePrincess
Paddle
提交
8232da7c
P
Paddle
项目概览
BaiXuePrincess
/
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
0
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
0
Issue
0
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
未验证
提交
8232da7c
编写于
9月 21, 2022
作者:
L
Leo Chen
提交者:
GitHub
9月 21, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
share threadpool of executor in dy2static (#46281)
上级
9e917a1e
变更
2
显示空白变更内容
内联
并排
Showing
2 changed file
with
34 addition
and
21 deletion
+34
-21
paddle/fluid/eager/to_static/run_program_op_node.h
paddle/fluid/eager/to_static/run_program_op_node.h
+33
-20
paddle/fluid/framework/new_executor/interpretercore.cc
paddle/fluid/framework/new_executor/interpretercore.cc
+1
-1
未找到文件。
paddle/fluid/eager/to_static/run_program_op_node.h
浏览文件 @
8232da7c
...
...
@@ -530,17 +530,34 @@ inline void RunProgramGradAPI(
auto
&
interpretercore_info_cache
=
paddle
::
framework
::
InterpreterCoreInfoCache
::
Instance
();
std
::
shared_ptr
<
paddle
::
framework
::
InterpreterCore
>
interpreter_core
=
nullptr
;
if
(
!
interpretercore_info_cache
.
Has
(
program_id
,
/*is_grad=*/
true
))
{
VLOG
(
2
)
<<
"No interpretercore cahce, so create a new interpretercore"
;
details
::
ShareTensorsIntoScope
(
out_grad
,
global_inner_scope
);
auto
interpreter_core
=
paddle
::
framework
::
CreateInterpreterCoreInfoToCache
(
interpreter_core
=
paddle
::
framework
::
CreateInterpreterCoreInfoToCache
(
*
backward_program
,
place
,
/*is_grad=*/
true
,
program_id
,
global_inner_scope
);
// share threadpool
// NOTE(zhiqiu): this only works interpreter_core is executed strictly
// after the related fwd_interpreter_core.
PADDLE_ENFORCE_EQ
(
interpretercore_info_cache
.
Has
(
program_id
,
false
),
true
,
paddle
::
platform
::
errors
::
NotFound
(
"The forward interpretercore of program %d is not found"
,
program_id
));
auto
fwd_interpreter_core
=
interpretercore_info_cache
.
GetMutable
(
program_id
,
/*is_grad=*/
false
)
.
core_
;
interpreter_core
->
ShareWorkQueueFrom
(
fwd_interpreter_core
);
VLOG
(
4
)
<<
"Share workqueue from "
<<
fwd_interpreter_core
.
get
()
<<
" to "
<<
interpreter_core
.
get
();
// get all eager gc vars
std
::
set
<
std
::
string
>
skip_eager_delete_vars
;
// all out_vars are skip_eager_var
...
...
@@ -552,17 +569,12 @@ inline void RunProgramGradAPI(
interpretercore_info_cache
.
UpdateSkipEagerDeleteVars
(
program_id
,
/*is_grad=*/
true
,
skip_eager_delete_vars
);
VLOG
(
2
)
<<
"Get skip GC vars size is: "
<<
skip_eager_delete_vars
.
size
();
if
(
backward_global_block
->
OpSize
()
>
0
)
{
// Debug info: scope info when run end
VLOG
(
3
)
<<
paddle
::
framework
::
GenScopeTreeDebugInfo
(
out_scope_vec
->
front
());
interpreter_core
->
Run
({});
}
}
else
{
VLOG
(
2
)
<<
"Get interpretercore cahce by program:"
<<
program_id
;
auto
&
cached_value
=
interpretercore_info_cache
.
GetMutable
(
program_id
,
/*is_grad=*/
true
);
auto
&
interpreter_core
=
cached_value
.
core_
;
interpreter_core
=
cached_value
.
core_
;
// update scope
details
::
ShareTensorsIntoScope
(
out_grad
,
global_inner_scope
);
if
(
interpreter_core
->
GetVariableScope
()
->
GetMutableScope
()
!=
...
...
@@ -572,6 +584,7 @@ inline void RunProgramGradAPI(
global_inner_scope
);
interpreter_core
->
reset_scope
(
global_inner_scope
);
}
}
if
(
backward_global_block
->
OpSize
()
>
0
)
{
// Debug info: scope info when run end
...
...
@@ -579,7 +592,7 @@ inline void RunProgramGradAPI(
out_scope_vec
->
front
());
interpreter_core
->
Run
({});
}
}
// Step 4. get outputs
details
::
ShareTensorsFromScopeWithPartialBlock
(
x_grad
,
*
forward_global_block
,
...
...
paddle/fluid/framework/new_executor/interpretercore.cc
浏览文件 @
8232da7c
...
...
@@ -284,7 +284,7 @@ void InterpreterCore::reset_scope(Scope* new_scope) {
void
InterpreterCore
::
ShareWorkQueueFrom
(
std
::
shared_ptr
<
InterpreterCore
>
src
)
{
async_work_queue_
=
src
->
GetWorkQueue
();
VLOG
(
8
)
<<
"Share AsyncWorkQueue from InterpreterCore("
<<
&
src
VLOG
(
8
)
<<
"Share AsyncWorkQueue from InterpreterCore("
<<
src
.
get
()
<<
") to InterpreterCore("
<<
this
<<
")"
;
}
...
...
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