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
2301_77200941
mindspore
提交
b7c7140b
M
mindspore
项目概览
2301_77200941
/
mindspore
与 Fork 源项目一致
Fork自
MindSpore / mindspore
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
M
mindspore
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
b7c7140b
编写于
8月 27, 2020
作者:
C
chenjianping
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
optimize gather,copy indices before parallel launch
上级
86c39d34
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
18 addition
and
11 deletion
+18
-11
mindspore/lite/src/runtime/kernel/arm/fp32/gather.cc
mindspore/lite/src/runtime/kernel/arm/fp32/gather.cc
+17
-10
mindspore/lite/src/runtime/kernel/arm/fp32/gather.h
mindspore/lite/src/runtime/kernel/arm/fp32/gather.h
+1
-1
未找到文件。
mindspore/lite/src/runtime/kernel/arm/fp32/gather.cc
浏览文件 @
b7c7140b
...
...
@@ -37,6 +37,10 @@ int GatherCPUKernel::Init() {
return
ReSize
();
}
GatherCPUKernel
::~
GatherCPUKernel
()
{
context_
->
allocator
->
Free
(
indices_data_
);
}
int
GatherCPUKernel
::
ReSize
()
{
return
RET_OK
;
}
int
GatherCPUKernel
::
DoGather
(
int
task_id
)
{
...
...
@@ -45,7 +49,6 @@ int GatherCPUKernel::DoGather(int task_id) {
auto
out_tensor
=
out_tensors_
.
at
(
0
);
auto
input_ptr
=
reinterpret_cast
<
float
*>
(
input_tensor
->
Data
());
auto
indices_ptr
=
reinterpret_cast
<
float
*>
(
indices_tensor
->
Data
());
auto
output_ptr
=
reinterpret_cast
<
float
*>
(
out_tensor
->
Data
());
auto
input_int32
=
reinterpret_cast
<
int32_t
*>
(
input_tensor
->
Data
());
...
...
@@ -57,13 +60,6 @@ int GatherCPUKernel::DoGather(int task_id) {
auto
axis
=
(
reinterpret_cast
<
GatherParameter
*>
(
op_parameter_
))
->
axis_
;
const
int
limit
=
in_shape
[
axis
];
for
(
int
i
=
0
;
i
<
indices_element_size
;
++
i
)
{
indices_data_
[
i
]
=
static_cast
<
int
>
(
indices_ptr
[
i
]);
if
(
indices_data_
[
i
]
>=
limit
)
{
MS_LOG
(
ERROR
)
<<
" indice data: "
<<
indices_data_
[
i
]
<<
" is not in [ 0, "
<<
limit
-
1
<<
" ]"
;
return
RET_ERROR
;
}
}
int
outer_size
=
1
,
inner_size
=
1
;
for
(
int
i
=
0
;
i
<
axis
;
++
i
)
{
...
...
@@ -106,12 +102,23 @@ int GatherCPUKernel::Run() {
}
auto
indices_tensor
=
in_tensors_
.
at
(
1
);
indices_data_
=
reinterpret_cast
<
int
*>
(
context_
->
allocator
->
Malloc
(
indices_tensor
->
ElementsNum
()
*
sizeof
(
int
)));
indices_data_
=
reinterpret_cast
<
int
*>
(
context_
->
allocator
->
Malloc
(
indices_tensor
->
Size
(
)));
if
(
indices_data_
==
nullptr
)
{
MS_LOG
(
ERROR
)
<<
"Memory allocation failed"
;
context_
->
allocator
->
Free
(
indices_data_
);
return
RET_ERROR
;
}
auto
in_shape
=
in_tensors_
.
at
(
0
)
->
shape
();
int
indices_element_size
=
indices_tensor
->
ElementsNum
();
auto
axis
=
(
reinterpret_cast
<
GatherParameter
*>
(
op_parameter_
))
->
axis_
;;
auto
indices_ptr
=
reinterpret_cast
<
float
*>
(
indices_tensor
->
Data
());
const
int
limit
=
in_shape
[
axis
];
for
(
int
i
=
0
;
i
<
indices_element_size
;
++
i
)
{
indices_data_
[
i
]
=
static_cast
<
int
>
(
indices_ptr
[
i
]);
if
(
indices_data_
[
i
]
>=
limit
)
{
MS_LOG
(
ERROR
)
<<
" indice data: "
<<
indices_data_
[
i
]
<<
" is not in [ 0, "
<<
limit
-
1
<<
" ]"
;
return
RET_ERROR
;
}
}
int
error_code
=
ParallelLaunch
(
THREAD_POOL_DEFAULT
,
GatherRun
,
this
,
op_parameter_
->
thread_num_
);
if
(
error_code
!=
RET_OK
)
{
MS_LOG
(
ERROR
)
<<
"Gather function error error_code["
<<
error_code
<<
"]"
;
...
...
mindspore/lite/src/runtime/kernel/arm/fp32/gather.h
浏览文件 @
b7c7140b
...
...
@@ -28,7 +28,7 @@ class GatherCPUKernel : public LiteKernel {
const
std
::
vector
<
lite
::
tensor
::
Tensor
*>
&
outputs
,
const
lite
::
Context
*
ctx
,
const
mindspore
::
lite
::
PrimitiveC
*
primitive
)
:
LiteKernel
(
parameter
,
inputs
,
outputs
,
ctx
,
primitive
)
{}
~
GatherCPUKernel
()
override
=
default
;
~
GatherCPUKernel
()
override
;
int
Init
()
override
;
int
ReSize
()
override
;
...
...
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