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
机器未来
Paddle
提交
8067a42b
P
Paddle
项目概览
机器未来
/
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 >>
提交
8067a42b
编写于
9月 18, 2017
作者:
L
Luo Tao
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
refine avg-pooling, which is exclusive. refine related code.
上级
59c48f98
变更
2
展开全部
隐藏空白更改
内联
并排
Showing
2 changed file
with
142 addition
and
172 deletion
+142
-172
paddle/cuda/src/hl_cuda_cnn.cu
paddle/cuda/src/hl_cuda_cnn.cu
+13
-18
paddle/math/Matrix.cpp
paddle/math/Matrix.cpp
+129
-154
未找到文件。
paddle/cuda/src/hl_cuda_cnn.cu
浏览文件 @
8067a42b
...
...
@@ -211,13 +211,11 @@ __global__ void KeAvgPoolForward(const int nthreads,
int
hstart
=
ph
*
strideH
-
padH
;
int
wstart
=
pw
*
strideW
-
padW
;
int
hend
=
min
(
hstart
+
sizeY
,
height
+
padH
);
int
wend
=
min
(
wstart
+
sizeX
,
width
+
padW
);
int
pool_size
=
(
hend
-
hstart
)
*
(
wend
-
wstart
);
int
hend
=
min
(
hstart
+
sizeY
,
height
);
int
wend
=
min
(
wstart
+
sizeX
,
width
);
hstart
=
max
(
hstart
,
0
);
wstart
=
max
(
wstart
,
0
);
hend
=
min
(
hend
,
height
);
wend
=
min
(
wend
,
width
);
int
pool_size
=
(
hend
-
hstart
)
*
(
wend
-
wstart
);
real
aveval
=
0
;
inputData
+=
(
frameNum
*
channels
+
c
)
*
height
*
width
;
...
...
@@ -299,11 +297,11 @@ __global__ void KeAvgPoolBackward(const int nthreads,
outGrad
+=
(
frameNum
*
outStride
+
offsetC
*
pooledH
*
pooledW
);
for
(
int
ph
=
phstart
;
ph
<
phend
;
++
ph
)
{
int
hstart
=
ph
*
strideH
-
padH
;
int
hend
=
min
(
hstart
+
sizeY
,
height
+
padH
);
for
(
int
pw
=
pwstart
;
pw
<
pwend
;
++
pw
)
{
// figure out the pooling size
int
hstart
=
ph
*
strideH
-
padH
;
int
wstart
=
pw
*
strideW
-
padW
;
int
hend
=
min
(
hstart
+
sizeY
,
height
+
padH
);
int
wend
=
min
(
wstart
+
sizeX
,
width
+
padW
);
int
poolsize
=
(
hend
-
hstart
)
*
(
wend
-
wstart
);
gradient
+=
outGrad
[
ph
*
pooledW
+
pw
]
/
poolsize
;
...
...
@@ -600,16 +598,13 @@ __global__ void KeAvgPool3DForward(const int nthreads,
int
dstart
=
pd
*
strideD
-
padD
;
int
hstart
=
ph
*
strideH
-
padH
;
int
wstart
=
pw
*
strideW
-
padW
;
int
dend
=
min
(
dstart
+
sizeZ
,
depth
+
padD
);
int
hend
=
min
(
hstart
+
sizeY
,
height
+
padH
);
int
wend
=
min
(
wstart
+
sizeX
,
width
+
padW
);
int
pool_size
=
(
dend
-
dstart
)
*
(
hend
-
hstart
)
*
(
wend
-
wstart
);
int
dend
=
min
(
dstart
+
sizeZ
,
depth
);
int
hend
=
min
(
hstart
+
sizeY
,
height
);
int
wend
=
min
(
wstart
+
sizeX
,
width
);
dstart
=
max
(
dstart
,
0
);
hstart
=
max
(
hstart
,
0
);
wstart
=
max
(
wstart
,
0
);
dend
=
min
(
dend
,
depth
);
hend
=
min
(
hend
,
height
);
wend
=
min
(
wend
,
width
);
int
pool_size
=
(
dend
-
dstart
)
*
(
hend
-
hstart
)
*
(
wend
-
wstart
);
real
aveval
=
0
;
inputData
+=
(
frameNum
*
channels
+
c
)
*
depth
*
height
*
width
;
...
...
@@ -712,14 +707,14 @@ __global__ void KeAvgPool3DBackward(const int nthreads,
outGrad
+=
(
frameNum
*
channels
+
offsetC
)
*
pooledD
*
pooledH
*
pooledW
;
for
(
int
pd
=
pdstart
;
pd
<
pdend
;
++
pd
)
{
int
dstart
=
pd
*
strideD
-
padD
;
int
dend
=
min
(
dstart
+
sizeZ
,
depth
+
padD
);
for
(
int
ph
=
phstart
;
ph
<
phend
;
++
ph
)
{
int
hstart
=
ph
*
strideH
-
padH
;
int
hend
=
min
(
hstart
+
sizeY
,
height
+
padH
);
for
(
int
pw
=
pwstart
;
pw
<
pwend
;
++
pw
)
{
// figure out the pooling size
int
dstart
=
pd
*
strideD
-
padD
;
int
hstart
=
ph
*
strideH
-
padH
;
int
wstart
=
pw
*
strideW
-
padW
;
int
dend
=
min
(
dstart
+
sizeZ
,
depth
+
padD
);
int
hend
=
min
(
hstart
+
sizeY
,
height
+
padH
);
int
wend
=
min
(
wstart
+
sizeX
,
width
+
padW
);
int
poolsize
=
(
dend
-
dstart
)
*
(
hend
-
hstart
)
*
(
wend
-
wstart
);
gradient
+=
outGrad
[(
pd
*
pooledH
+
ph
)
*
pooledW
+
pw
]
/
poolsize
;
...
...
paddle/math/Matrix.cpp
浏览文件 @
8067a42b
此差异已折叠。
点击以展开。
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