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
s920243400
PaddleDetection
提交
920a9609
P
PaddleDetection
项目概览
s920243400
/
PaddleDetection
与 Fork 源项目一致
Fork自
PaddlePaddle / PaddleDetection
通知
2
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
PaddleDetection
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
提交
920a9609
编写于
4月 12, 2019
作者:
Q
Qiao Longfei
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
optimize merge add if input rows of all selected rows is not duplicated
上级
a06f4b2b
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
56 addition
and
26 deletion
+56
-26
paddle/fluid/operators/math/selected_rows_functor.cc
paddle/fluid/operators/math/selected_rows_functor.cc
+56
-26
未找到文件。
paddle/fluid/operators/math/selected_rows_functor.cc
浏览文件 @
920a9609
...
...
@@ -296,6 +296,7 @@ struct MergeAdd<platform::CPUDeviceContext, T> {
auto
input_height
=
has_value_input
->
height
();
framework
::
SelectedRows
&
out
=
*
output
;
std
::
set
<
int64_t
>
merged_row_set
;
size_t
row_num
=
0
;
for
(
auto
*
input
:
inputs
)
{
if
(
input
->
rows
().
size
()
==
0
)
{
continue
;
...
...
@@ -305,42 +306,71 @@ struct MergeAdd<platform::CPUDeviceContext, T> {
"dimension except for the first one"
);
PADDLE_ENFORCE_EQ
(
input_height
,
input
->
height
(),
"all input should have same height"
);
row_num
+=
input
->
rows
().
size
();
merged_row_set
.
insert
(
input
->
rows
().
begin
(),
input
->
rows
().
end
());
}
std
::
vector
<
int64_t
>
merge_rows
(
merged_row_set
.
begin
(),
merged_row_set
.
end
());
if
(
sorted_result
)
{
std
::
sort
(
merge_rows
.
begin
(),
merge_rows
.
end
());
}
std
::
unordered_map
<
int64_t
,
size_t
>
rows_to_id
;
for
(
size_t
i
=
0
;
i
<
merge_rows
.
size
();
++
i
)
{
rows_to_id
[
merge_rows
[
i
]]
=
i
;
}
out
.
set_rows
(
merge_rows
);
out
.
set_height
(
input_height
);
out
.
mutable_value
()
->
mutable_data
<
T
>
(
framework
::
make_ddim
(
{
static_cast
<
int64_t
>
(
merge
_rows
.
size
()),
input_width
}),
{
static_cast
<
int64_t
>
(
merge
d_row_set
.
size
()),
input_width
}),
context
.
GetPlace
());
auto
*
out_data
=
out
.
mutable_value
()
->
data
<
T
>
();
math
::
SetConstant
<
platform
::
CPUDeviceContext
,
T
>
constant_functor
;
constant_functor
(
context
,
out
.
mutable_value
(),
0.0
);
if
(
merged_row_set
.
size
()
==
row_num
&&
!
sorted_result
)
{
// no duplicated ids, just concat the result together
std
::
vector
<
int64_t
>
merge_rows
;
merge_rows
.
reserve
(
row_num
);
// concat rows
for
(
auto
*
in
:
inputs
)
{
merge_rows
.
insert
(
merge_rows
.
end
(),
in
->
rows
().
begin
(),
in
->
rows
().
end
());
}
out
.
set_rows
(
merge_rows
);
auto
in_place
=
inputs
[
0
]
->
place
();
auto
out_place
=
out
.
place
();
int64_t
copied_numel
=
0
;
for
(
auto
*
in
:
inputs
)
{
auto
*
in_data
=
in
->
value
().
data
<
T
>
();
auto
in_numel
=
in
->
value
().
numel
();
memory
::
Copy
(
boost
::
get
<
platform
::
CPUPlace
>
(
out_place
),
out_data
+
copied_numel
,
boost
::
get
<
platform
::
CPUPlace
>
(
in_place
),
in_data
,
in_numel
*
sizeof
(
T
));
copied_numel
+=
in_numel
;
}
}
else
{
std
::
vector
<
int64_t
>
merge_rows
(
merged_row_set
.
begin
(),
merged_row_set
.
end
());
auto
*
out_data
=
out
.
mutable_value
()
->
data
<
T
>
();
if
(
sorted_result
)
{
std
::
sort
(
merge_rows
.
begin
(),
merge_rows
.
end
());
}
auto
blas
=
math
::
GetBlas
<
platform
::
CPUDeviceContext
,
T
>
(
context
);
for
(
auto
*
input
:
inputs
)
{
if
(
input
->
rows
().
size
()
==
0
)
{
continue
;
out
.
set_rows
(
merge_rows
);
math
::
SetConstant
<
platform
::
CPUDeviceContext
,
T
>
constant_functor
;
constant_functor
(
context
,
out
.
mutable_value
(),
0.0
);
std
::
unordered_map
<
int64_t
,
size_t
>
rows_to_id
;
for
(
size_t
i
=
0
;
i
<
merge_rows
.
size
();
++
i
)
{
rows_to_id
[
merge_rows
[
i
]]
=
i
;
}
auto
*
input_data
=
input
->
value
().
data
<
T
>
();
auto
&
input_rows
=
input
->
rows
();
for
(
size_t
i
=
0
;
i
<
input_rows
.
size
();
i
++
)
{
size_t
out_i
=
rows_to_id
[
input_rows
[
i
]];
elementwise_add_to
<
platform
::
CPUDeviceContext
,
T
>
(
context
,
&
blas
,
static_cast
<
size_t
>
(
input_width
),
&
input_data
[
i
*
input_width
],
&
out_data
[
out_i
*
input_width
]);
auto
blas
=
math
::
GetBlas
<
platform
::
CPUDeviceContext
,
T
>
(
context
);
for
(
auto
*
input
:
inputs
)
{
if
(
input
->
rows
().
size
()
==
0
)
{
continue
;
}
auto
*
input_data
=
input
->
value
().
data
<
T
>
();
auto
&
input_rows
=
input
->
rows
();
for
(
size_t
i
=
0
;
i
<
input_rows
.
size
();
i
++
)
{
size_t
out_i
=
rows_to_id
[
input_rows
[
i
]];
elementwise_add_to
<
platform
::
CPUDeviceContext
,
T
>
(
context
,
&
blas
,
static_cast
<
size_t
>
(
input_width
),
&
input_data
[
i
*
input_width
],
&
out_data
[
out_i
*
input_width
]);
}
}
}
}
...
...
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