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
PaddlePaddle
PaddleDetection
提交
953f8ddf
P
PaddleDetection
项目概览
PaddlePaddle
/
PaddleDetection
大约 1 年 前同步成功
通知
695
Star
11112
Fork
2696
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
184
列表
看板
标记
里程碑
合并请求
40
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
184
Issue
184
列表
看板
标记
里程碑
合并请求
40
合并请求
40
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
953f8ddf
编写于
8月 01, 2017
作者:
H
hedaoyuan
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Support groups in NNPACKFunction.
上级
0973c2c9
变更
1
显示空白变更内容
内联
并排
Showing
1 changed file
with
53 addition
and
47 deletion
+53
-47
paddle/function/nnpack/NNPACKConvOp.cpp
paddle/function/nnpack/NNPACKConvOp.cpp
+53
-47
未找到文件。
paddle/function/nnpack/NNPACKConvOp.cpp
浏览文件 @
953f8ddf
...
@@ -49,9 +49,7 @@ class NNPACKConvFunction : public ConvFunctionBase {
...
@@ -49,9 +49,7 @@ class NNPACKConvFunction : public ConvFunctionBase {
public:
public:
void
init
(
const
FuncConfig
&
config
)
override
{
void
init
(
const
FuncConfig
&
config
)
override
{
ConvFunctionBase
::
init
(
config
);
ConvFunctionBase
::
init
(
config
);
CHECK_EQ
(
groups_
,
(
size_t
)
1
);
algorithm_
=
get_nnp_convolution_algorithm
(
config
.
get
<
std
::
string
>
(
"algo"
));
algorithm_
=
get_nnp_convolution_algorithm
(
config
.
get
<
std
::
string
>
(
"algo"
));
// algorithm_ = nnp_convolution_algorithm_auto;
transform_strategy_
=
nnp_convolution_transform_strategy_compute
;
transform_strategy_
=
nnp_convolution_transform_strategy_compute
;
nnp_status
status
=
nnp_initialize
();
nnp_status
status
=
nnp_initialize
();
CHECK_EQ
(
status
,
nnp_status_success
);
CHECK_EQ
(
status
,
nnp_status_success
);
...
@@ -67,8 +65,7 @@ public:
...
@@ -67,8 +65,7 @@ public:
}
}
}
}
virtual
void
check
(
const
BufferArgs
&
inputs
,
void
check
(
const
BufferArgs
&
inputs
,
const
BufferArgs
&
outputs
)
override
{
const
BufferArgs
&
outputs
)
override
{
const
TensorShape
&
input
=
inputs
[
0
].
shape
();
const
TensorShape
&
input
=
inputs
[
0
].
shape
();
const
TensorShape
&
filter
=
inputs
[
1
].
shape
();
const
TensorShape
&
filter
=
inputs
[
1
].
shape
();
const
TensorShape
&
output
=
outputs
[
0
].
shape
();
const
TensorShape
&
output
=
outputs
[
0
].
shape
();
...
@@ -91,8 +88,8 @@ public:
...
@@ -91,8 +88,8 @@ public:
size_t
filterHeight
=
getFilterHeight
(
filter
);
size_t
filterHeight
=
getFilterHeight
(
filter
);
size_t
filterWidth
=
getFilterWidth
(
filter
);
size_t
filterWidth
=
getFilterWidth
(
filter
);
size_t
outputChannels
=
output
[
1
];
size_t
outputChannels
=
output
[
1
];
//
size_t outputHeight = output[2];
size_t
outputHeight
=
output
[
2
];
//
size_t outputWidth = output[3];
size_t
outputWidth
=
output
[
3
];
nnp_size
inputSize
=
{.
width
=
inputWidth
,
.
height
=
inputHeight
};
nnp_size
inputSize
=
{.
width
=
inputWidth
,
.
height
=
inputHeight
};
nnp_padding
padding
=
{.
top
=
(
size_t
)
paddingH
(),
nnp_padding
padding
=
{.
top
=
(
size_t
)
paddingH
(),
...
@@ -171,20 +168,25 @@ public:
...
@@ -171,20 +168,25 @@ public:
}
}
}
}
size_t
inputOffset
=
inputChannels
/
groups_
*
inputHeight
*
inputWidth
;
size_t
outputOffset
=
outputChannels
/
groups_
*
outputHeight
*
outputWidth
;
size_t
filterOffset
=
filter
.
getElements
()
/
groups_
;
if
(
batchSize
==
1
)
{
if
(
batchSize
==
1
)
{
for
(
size_t
g
=
0
;
g
<
groups_
;
g
++
)
{
nnp_status
status
=
nnp_status
status
=
nnp_convolution_inference
(
algorithm_
,
nnp_convolution_inference
(
algorithm_
,
transform_strategy_
,
transform_strategy_
,
inputChannels
,
inputChannels
/
groups_
,
outputChannels
,
outputChannels
/
groups_
,
inputSize
,
inputSize
,
padding
,
padding
,
kernelSize
,
kernelSize
,
outputSubsampling
,
outputSubsampling
,
inputData
,
inputData
+
inputOffset
*
g
,
filterData
,
filterData
+
filterOffset
*
g
,
nullptr
,
/* bias */
nullptr
,
/* bias */
outputData
,
outputData
+
outputOffset
*
g
,
bufferPtr
,
bufferPtr
,
sizePtr
,
sizePtr
,
nnp_activation_identity
,
nnp_activation_identity
,
...
@@ -192,21 +194,24 @@ public:
...
@@ -192,21 +194,24 @@ public:
threadpool_
,
/* threadpool */
threadpool_
,
/* threadpool */
nullptr
);
nullptr
);
CHECK_EQ
(
status
,
nnp_status_success
);
CHECK_EQ
(
status
,
nnp_status_success
);
}
}
else
{
}
else
{
for
(
size_t
g
=
0
;
g
<
groups_
;
g
++
)
{
// only supports stride = 1
// only supports stride = 1
CHECK_EQ
(
strideH
(),
1
);
CHECK_EQ
(
strideH
(),
1
);
CHECK_EQ
(
strideW
(),
1
);
CHECK_EQ
(
strideW
(),
1
);
nnp_status
status
=
nnp_convolution_output
(
algorithm_
,
nnp_status
status
=
nnp_convolution_output
(
algorithm_
,
batchSize
,
batchSize
,
inputChannels
,
inputChannels
/
groups_
,
outputChannels
,
outputChannels
/
groups_
,
inputSize
,
inputSize
,
padding
,
padding
,
kernelSize
,
kernelSize
,
inputData
,
inputData
+
inputOffset
*
g
,
filterData
,
filterData
+
filterOffset
*
g
,
nullptr
,
/* bias */
nullptr
,
/* bias */
outputData
,
outputData
+
outputOffset
*
g
,
bufferPtr
,
bufferPtr
,
sizePtr
,
sizePtr
,
nnp_activation_identity
,
nnp_activation_identity
,
...
@@ -216,6 +221,7 @@ public:
...
@@ -216,6 +221,7 @@ public:
CHECK_EQ
(
status
,
nnp_status_success
);
CHECK_EQ
(
status
,
nnp_status_success
);
}
}
}
}
}
static
void
create_nnpack_threadpool
()
{
static
void
create_nnpack_threadpool
()
{
if
(
FLAGS_nnpack_num_threads
&&
threadpool_
==
nullptr
)
{
if
(
FLAGS_nnpack_num_threads
&&
threadpool_
==
nullptr
)
{
...
...
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