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
wmsofts
Paddle
提交
83053221
P
Paddle
项目概览
wmsofts
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
0
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
体验新版 GitCode,发现更多精彩内容 >>
提交
83053221
编写于
5月 09, 2018
作者:
C
chengduoZH
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
extract method from broadcast::RunImpl
上级
93368aac
变更
5
隐藏空白更改
内联
并排
Showing
5 changed file
with
56 addition
and
33 deletion
+56
-33
paddle/fluid/framework/details/broadcast_op_handle.cc
paddle/fluid/framework/details/broadcast_op_handle.cc
+35
-21
paddle/fluid/framework/details/broadcast_op_handle.h
paddle/fluid/framework/details/broadcast_op_handle.h
+3
-0
paddle/fluid/framework/details/op_handle_base.h
paddle/fluid/framework/details/op_handle_base.h
+6
-2
paddle/fluid/framework/details/reduce_op_handle.cc
paddle/fluid/framework/details/reduce_op_handle.cc
+11
-9
paddle/fluid/framework/details/send_op_handle.cc
paddle/fluid/framework/details/send_op_handle.cc
+1
-1
未找到文件。
paddle/fluid/framework/details/broadcast_op_handle.cc
浏览文件 @
83053221
...
...
@@ -48,29 +48,9 @@ void BroadcastOpHandle::RunImpl() {
auto
*
in_var
=
var_scopes
.
at
(
in_var_handle
->
scope_idx_
)
->
FindVar
(
in_var_handle
->
name_
);
PADDLE_ENFORCE_NOT_NULL
(
in_var
);
Tensor
&
in_tensor
=
VariableVisitor
::
GetMutableTensor
(
in_var
);
// NOTE: The tensors' Place of input and output must be all on GPU or all on
// CPU.
for
(
auto
*
out_var_handle
:
out_var_handles
)
{
if
(
out_var_handle
->
IsTheSameVar
(
*
in_var_handle
))
{
continue
;
}
auto
t_out_p
=
out_var_handle
->
place_
;
auto
*
out_var
=
var_scopes
.
at
(
out_var_handle
->
scope_idx_
)
->
FindVar
(
out_var_handle
->
name_
);
PADDLE_ENFORCE_NOT_NULL
(
out_var
);
if
(
platform
::
is_gpu_place
(
in_tensor
.
place
()))
{
PADDLE_ENFORCE
(
platform
::
is_gpu_place
(
t_out_p
),
"Places of input and output must be all on GPU."
);
}
else
{
t_out_p
=
platform
::
CPUPlace
();
}
VariableVisitor
::
ShareDimsAndLoD
(
*
in_var
,
out_var
);
VariableVisitor
::
GetMutableTensor
(
out_var
).
mutable_data
(
t_out_p
,
in_tensor
.
type
());
}
InitOutputValue
(
*
in_var_handle
,
out_var_handles
);
if
(
platform
::
is_cpu_place
(
in_tensor
.
place
()))
{
for
(
auto
*
out_var_handle
:
out_var_handles
)
{
...
...
@@ -145,6 +125,40 @@ void BroadcastOpHandle::RunImpl() {
}
}
void
BroadcastOpHandle
::
InitOutputValue
(
const
VarHandle
&
in_var_handle
,
const
std
::
vector
<
VarHandle
*>
&
out_var_handles
)
const
{
std
::
vector
<
const
Scope
*>
var_scopes
;
for
(
auto
*
s
:
local_scopes_
)
{
var_scopes
.
emplace_back
(
s
->
FindVar
(
kLocalExecScopeName
)
->
Get
<
Scope
*>
());
}
auto
*
in_var
=
var_scopes
.
at
(
in_var_handle
.
scope_idx_
)
->
FindVar
(
in_var_handle
.
name_
);
Tensor
&
in_tensor
=
VariableVisitor
::
GetMutableTensor
(
in_var
);
// NOTE: The tensors' Place of input and output must be all on GPU or all on
// CPU.
for
(
auto
*
out_var_handle
:
out_var_handles
)
{
if
(
out_var_handle
->
IsTheSameVar
(
in_var_handle
))
{
continue
;
}
auto
t_out_p
=
out_var_handle
->
place_
;
auto
*
out_var
=
var_scopes
.
at
(
out_var_handle
->
scope_idx_
)
->
FindVar
(
out_var_handle
->
name_
);
PADDLE_ENFORCE_NOT_NULL
(
out_var
);
if
(
is_gpu_place
(
in_tensor
.
place
()))
{
PADDLE_ENFORCE
(
platform
::
is_gpu_place
(
t_out_p
),
"Places of input and output must be all on GPU."
);
}
else
{
t_out_p
=
platform
::
CPUPlace
();
}
VariableVisitor
::
ShareDimsAndLoD
(
*
in_var
,
out_var
);
VariableVisitor
::
GetMutableTensor
(
out_var
).
mutable_data
(
t_out_p
,
in_tensor
.
type
());
}
}
std
::
string
BroadcastOpHandle
::
Name
()
const
{
return
"broadcast"
;
}
}
// namespace details
}
// namespace framework
...
...
paddle/fluid/framework/details/broadcast_op_handle.h
浏览文件 @
83053221
...
...
@@ -64,6 +64,9 @@ struct BroadcastOpHandle : public OpHandleBase {
#ifdef PADDLE_WITH_CUDA
const
platform
::
NCCLContextMap
*
nccl_ctxs_
;
#endif
void
InitOutputValue
(
const
VarHandle
&
in_var_handle
,
const
std
::
vector
<
VarHandle
*>
&
out_var_handles
)
const
;
};
}
// namespace details
}
// namespace framework
...
...
paddle/fluid/framework/details/op_handle_base.h
浏览文件 @
83053221
...
...
@@ -44,10 +44,14 @@ class OpHandleBase {
void
AddOutput
(
VarHandleBase
*
out
);
// Wait inputs are generated, this Wait is asynchronous operation.
// This method adds the wait events of all the input on all the device
// context.
// NODE: This Wait is asynchronous operation.
virtual
void
WaitInputVarGenerated
();
// Wait inputs are generated, this Wait is asynchronous operation.
// This method adds the wait events of all the input on the specified device
// context.
// NODE: This Wait is asynchronous operation.
virtual
void
WaitInputVarGenerated
(
const
platform
::
Place
&
place
);
virtual
bool
NeedWait
(
VarHandleBase
*
in_var
);
...
...
paddle/fluid/framework/details/reduce_op_handle.cc
浏览文件 @
83053221
...
...
@@ -80,19 +80,21 @@ void ReduceOpHandle::RunImpl() {
}
if
(
pre_in_var
->
IsType
<
framework
::
SelectedRows
>
())
{
std
::
vector
<
const
SelectedRows
*>
in_selected_rows
=
GetInputValues
<
SelectedRows
>
(
in_var_handles
,
var_scopes
);
GatherSelectedRows
(
in_selected_rows
,
in_places
,
dev_ctxes_
,
t_out_p
,
out_var
->
GetMutable
<
framework
::
SelectedRows
>
());
this
->
RunAndRecordEvent
([
&
]
{
std
::
vector
<
const
SelectedRows
*>
in_selected_rows
=
GetInputValues
<
SelectedRows
>
(
in_var_handles
,
var_scopes
);
GatherSelectedRows
(
in_selected_rows
,
in_places
,
dev_ctxes_
,
t_out_p
,
out_var
->
GetMutable
<
framework
::
SelectedRows
>
());
});
}
else
{
std
::
vector
<
const
LoDTensor
*>
lod_tensors
=
GetInputValues
<
LoDTensor
>
(
in_var_handles
,
var_scopes
);
if
(
paddle
::
platform
::
is_cpu_place
(
lod_tensors
[
0
]
->
place
()))
{
ReduceLoDTensor
func
(
lod_tensors
,
out_var
->
GetMutable
<
framework
::
LoDTensor
>
());
VisitDataType
(
ToDataType
(
lod_tensors
[
0
]
->
type
()),
func
);
this
->
RunAndRecordEvent
([
&
]
{
ReduceLoDTensor
func
(
lod_tensors
,
out_var
->
GetMutable
<
framework
::
LoDTensor
>
());
VisitDataType
(
ToDataType
(
lod_tensors
[
0
]
->
type
()),
func
);
});
}
else
if
(
paddle
::
platform
::
is_gpu_place
(
lod_tensors
[
0
]
->
place
()))
{
#ifdef PADDLE_WITH_CUDA
auto
pre_in
=
pre_in_var
->
Get
<
framework
::
LoDTensor
>
();
...
...
paddle/fluid/framework/details/send_op_handle.cc
浏览文件 @
83053221
...
...
@@ -27,7 +27,7 @@ SendOpHandle::SendOpHandle(const framework::OpDesc &op_desc,
void
SendOpHandle
::
RunImpl
()
{
// Wait input done
WaitInputVarGenerated
();
WaitInputVarGenerated
(
place_
);
auto
&
tmp_scope
=
local_scope_
->
FindVar
(
kLocalExecScopeName
)
->
Get
<
Scope
*>
();
// FIXME(wuyi): can not use RunAndRecordEvent here, for it will cause dead
// lock.
...
...
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