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
wzpei9
Paddle
提交
2d0e5592
P
Paddle
项目概览
wzpei9
/
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 >>
未验证
提交
2d0e5592
编写于
7月 03, 2018
作者:
Y
yuyang18
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Use std::map for Place <--> DeviceContext
上级
f7fd711e
变更
6
隐藏空白更改
内联
并排
Showing
6 changed file
with
13 addition
and
32 deletion
+13
-32
paddle/fluid/framework/details/op_handle_base.cc
paddle/fluid/framework/details/op_handle_base.cc
+2
-8
paddle/fluid/framework/details/op_handle_base.h
paddle/fluid/framework/details/op_handle_base.h
+2
-4
paddle/fluid/framework/details/reduce_and_gather.h
paddle/fluid/framework/details/reduce_and_gather.h
+1
-2
paddle/fluid/platform/device_context.cc
paddle/fluid/platform/device_context.cc
+2
-1
paddle/fluid/platform/device_context.h
paddle/fluid/platform/device_context.h
+3
-5
paddle/fluid/platform/place.h
paddle/fluid/platform/place.h
+3
-12
未找到文件。
paddle/fluid/framework/details/op_handle_base.cc
浏览文件 @
2d0e5592
...
...
@@ -124,16 +124,10 @@ void OpHandleBase::RunAndRecordEvent(const std::function<void()> &callback) {
#ifdef PADDLE_WITH_CUDA
if
(
!
events_
.
empty
())
{
// Use event
std
::
function
<
void
()
>
method
=
callback
;
// NOTE(zcd): device context must be ordered here because RecordEvent
// will use a mutex to ensure the safe of multi-threads.
std
::
map
<
platform
::
DeviceContext
*
,
platform
::
Place
>
ordered_ctxes
;
for
(
auto
&
p
:
dev_ctxes_
)
{
ordered_ctxes
.
emplace
(
p
.
second
,
p
.
first
);
}
for
(
auto
&
p
:
ordered_ctxes
)
{
method
=
[
method
,
p
,
this
]()
{
static_cast
<
platform
::
CUDADeviceContext
*>
(
p
.
first
)
->
RecordEvent
(
events_
.
at
(
boost
::
get
<
platform
::
CUDAPlace
>
(
p
.
second
).
device
),
static_cast
<
platform
::
CUDADeviceContext
*>
(
p
.
second
)
->
RecordEvent
(
events_
.
at
(
boost
::
get
<
platform
::
CUDAPlace
>
(
p
.
first
).
device
),
method
);
};
}
...
...
paddle/fluid/framework/details/op_handle_base.h
浏览文件 @
2d0e5592
...
...
@@ -13,9 +13,9 @@
// limitations under the License.
#pragma once
#include <map>
#include <string>
#include <vector>
#include "paddle/fluid/framework/details/var_handle.h"
#include "paddle/fluid/platform/device_context.h"
#include "paddle/fluid/platform/macros.h"
...
...
@@ -92,9 +92,7 @@ class OpHandleBase {
std
::
vector
<
VarHandleBase
*>
inputs_
;
std
::
vector
<
VarHandleBase
*>
outputs_
;
std
::
unordered_map
<
platform
::
Place
,
platform
::
DeviceContext
*
,
platform
::
PlaceHash
>
dev_ctxes_
;
std
::
map
<
platform
::
Place
,
platform
::
DeviceContext
*>
dev_ctxes_
;
#ifdef PADDLE_WITH_CUDA
std
::
unordered_map
<
int
,
cudaEvent_t
>
events_
;
...
...
paddle/fluid/framework/details/reduce_and_gather.h
浏览文件 @
2d0e5592
...
...
@@ -54,8 +54,7 @@ struct ReduceLoDTensor {
inline
void
GatherSelectedRows
(
const
std
::
vector
<
const
SelectedRows
*>
&
src_selecte_rows_
,
const
std
::
vector
<
platform
::
Place
>
&
in_places
,
const
std
::
unordered_map
<
platform
::
Place
,
platform
::
DeviceContext
*
,
platform
::
PlaceHash
>
&
dev_ctxes
,
const
std
::
map
<
platform
::
Place
,
platform
::
DeviceContext
*>
&
dev_ctxes
,
const
platform
::
Place
&
out_place
,
SelectedRows
*
dst_selecte_rows
)
{
PADDLE_ENFORCE
(
!
src_selecte_rows_
.
empty
());
...
...
paddle/fluid/platform/device_context.cc
浏览文件 @
2d0e5592
...
...
@@ -10,6 +10,7 @@ See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/fluid/platform/device_context.h"
#include <set>
#include <string>
#include <unordered_set>
#include <vector>
...
...
@@ -35,7 +36,7 @@ DeviceContextPool::DeviceContextPool(
const
std
::
vector
<
platform
::
Place
>&
places
)
{
PADDLE_ENFORCE_GT
(
places
.
size
(),
0
);
using
PtrType
=
std
::
unique_ptr
<
DeviceContext
>
;
std
::
unordered_set
<
Place
,
PlaceHash
>
set
;
std
::
set
<
Place
>
set
;
for
(
auto
&
p
:
places
)
{
set
.
insert
(
p
);
}
...
...
paddle/fluid/platform/device_context.h
浏览文件 @
2d0e5592
...
...
@@ -27,12 +27,12 @@ limitations under the License. */
#include <mkldnn.hpp>
#endif
#include <map>
#include "glog/logging.h"
#include "paddle/fluid/platform/enforce.h"
#include "paddle/fluid/platform/place.h"
#include "unsupported/Eigen/CXX11/Tensor"
#include "glog/logging.h"
namespace
paddle
{
namespace
platform
{
...
...
@@ -201,9 +201,7 @@ class DeviceContextPool {
private:
static
DeviceContextPool
*
pool
;
std
::
unordered_map
<
const
platform
::
Place
,
std
::
unique_ptr
<
platform
::
DeviceContext
>
,
PlaceHash
>
device_contexts_
;
std
::
map
<
Place
,
std
::
unique_ptr
<
DeviceContext
>>
device_contexts_
;
DISABLE_COPY_AND_ASSIGN
(
DeviceContextPool
);
};
...
...
paddle/fluid/platform/place.h
浏览文件 @
2d0e5592
...
...
@@ -30,6 +30,7 @@ struct CPUPlace {
// needed for variant equality comparison
inline
bool
operator
==
(
const
CPUPlace
&
)
const
{
return
true
;
}
inline
bool
operator
!=
(
const
CPUPlace
&
)
const
{
return
false
;
}
inline
bool
operator
<
(
const
CPUPlace
&
)
const
{
return
false
;
}
};
struct
CUDAPlace
{
...
...
@@ -42,6 +43,7 @@ struct CUDAPlace {
return
device
==
o
.
device
;
}
inline
bool
operator
!=
(
const
CUDAPlace
&
o
)
const
{
return
!
(
*
this
==
o
);
}
inline
bool
operator
<
(
const
CUDAPlace
&
o
)
const
{
return
device
<
o
.
device
;
}
int
device
;
};
...
...
@@ -52,6 +54,7 @@ struct CUDAPinnedPlace {
// needed for variant equality comparison
inline
bool
operator
==
(
const
CUDAPinnedPlace
&
)
const
{
return
true
;
}
inline
bool
operator
!=
(
const
CUDAPinnedPlace
&
)
const
{
return
false
;
}
inline
bool
operator
<
(
const
CUDAPinnedPlace
&
)
const
{
return
false
;
}
};
struct
IsCUDAPlace
:
public
boost
::
static_visitor
<
bool
>
{
...
...
@@ -89,18 +92,6 @@ bool is_cuda_pinned_place(const Place &);
bool
places_are_same_class
(
const
Place
&
,
const
Place
&
);
bool
is_same_place
(
const
Place
&
,
const
Place
&
);
struct
PlaceHash
{
std
::
size_t
operator
()(
const
Place
&
p
)
const
{
constexpr
size_t
num_dev_bits
=
4
;
std
::
hash
<
int
>
ihash
;
size_t
dev_id
=
0
;
if
(
is_gpu_place
(
p
))
{
dev_id
=
boost
::
get
<
CUDAPlace
>
(
p
).
device
;
}
return
ihash
(
dev_id
<<
num_dev_bits
|
p
.
which
());
}
};
std
::
ostream
&
operator
<<
(
std
::
ostream
&
,
const
Place
&
);
template
<
typename
Visitor
>
...
...
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