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
夏有凉风,冬有雪
Paddle
提交
74912c7d
P
Paddle
项目概览
夏有凉风,冬有雪
/
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
74912c7d
编写于
11月 15, 2017
作者:
C
chengduoZH
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
fix data layout
上级
c5330fa7
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
20 addition
and
13 deletion
+20
-13
paddle/operators/conv_transpose_cudnn_op.cu
paddle/operators/conv_transpose_cudnn_op.cu
+13
-7
paddle/platform/cudnn_helper.h
paddle/platform/cudnn_helper.h
+7
-6
未找到文件。
paddle/operators/conv_transpose_cudnn_op.cu
浏览文件 @
74912c7d
...
@@ -54,15 +54,21 @@ class CudnnConvTransposeOpKernel : public framework::OpKernel<T> {
...
@@ -54,15 +54,21 @@ class CudnnConvTransposeOpKernel : public framework::OpKernel<T> {
ScopedTensorDescriptor
output_desc
;
ScopedTensorDescriptor
output_desc
;
ScopedFilterDescriptor
filter_desc
;
ScopedFilterDescriptor
filter_desc
;
ScopedConvolutionDescriptor
conv_desc
;
ScopedConvolutionDescriptor
conv_desc
;
DataLayout
layout
=
DataLayout
::
kNCHW
;
DataLayout
layout
;
if
(
strides
.
size
()
==
2U
)
{
layout
=
DataLayout
::
kNCHW
;
}
else
{
layout
=
DataLayout
::
kNCDHW
;
}
//
N, M, H, W
//
(N, M, H, W) or (N, M, D, H, W)
cudnnTensorDescriptor_t
cudnn_input_desc
=
input_desc
.
descriptor
<
T
>
(
cudnnTensorDescriptor_t
cudnn_input_desc
=
input_desc
.
descriptor
<
T
>
(
layout
,
framework
::
vectorize2int
(
input
->
dims
()));
layout
,
framework
::
vectorize2int
(
input
->
dims
()));
//
N, C, O_h, O_w
//
(N, C, O_h, O_w) or (N, C, O_d, O_h, O_w)
cudnnTensorDescriptor_t
cudnn_output_desc
=
output_desc
.
descriptor
<
T
>
(
cudnnTensorDescriptor_t
cudnn_output_desc
=
output_desc
.
descriptor
<
T
>
(
layout
,
framework
::
vectorize2int
(
output
->
dims
()));
layout
,
framework
::
vectorize2int
(
output
->
dims
()));
//
M, C, K_h, K_w
//
(M, C, K_h, K_w) or (M, C, K_d, K_h, K_w)
cudnnFilterDescriptor_t
cudnn_filter_desc
=
filter_desc
.
descriptor
<
T
>
(
cudnnFilterDescriptor_t
cudnn_filter_desc
=
filter_desc
.
descriptor
<
T
>
(
layout
,
framework
::
vectorize2int
(
filter
->
dims
()));
layout
,
framework
::
vectorize2int
(
filter
->
dims
()));
cudnnConvolutionDescriptor_t
cudnn_conv_desc
=
cudnnConvolutionDescriptor_t
cudnn_conv_desc
=
...
@@ -136,13 +142,13 @@ class CudnnConvTransposeGradOpKernel : public framework::OpKernel<T> {
...
@@ -136,13 +142,13 @@ class CudnnConvTransposeGradOpKernel : public framework::OpKernel<T> {
ScopedConvolutionDescriptor
conv_desc
;
ScopedConvolutionDescriptor
conv_desc
;
DataLayout
layout
=
DataLayout
::
kNCHW
;
DataLayout
layout
=
DataLayout
::
kNCHW
;
// Input: (N, M, H, W)
// Input: (N, M, H, W)
or (N, M, D, H, W)
cudnnTensorDescriptor_t
cudnn_input_desc
=
input_desc
.
descriptor
<
T
>
(
cudnnTensorDescriptor_t
cudnn_input_desc
=
input_desc
.
descriptor
<
T
>
(
layout
,
framework
::
vectorize2int
(
input
->
dims
()));
layout
,
framework
::
vectorize2int
(
input
->
dims
()));
// Output: (N, C, O_
H, O_W
)
// Output: (N, C, O_
h, O_w) or (N, C, O_d, O_h, O_w
)
cudnnTensorDescriptor_t
cudnn_output_desc
=
output_desc
.
descriptor
<
T
>
(
cudnnTensorDescriptor_t
cudnn_output_desc
=
output_desc
.
descriptor
<
T
>
(
layout
,
framework
::
vectorize2int
(
output_grad
->
dims
()));
layout
,
framework
::
vectorize2int
(
output_grad
->
dims
()));
// Filter (M, C, K_
H, K_W
)
// Filter (M, C, K_
h, K_w) or (M, C, K_d K_h, K_w
)
cudnnFilterDescriptor_t
cudnn_filter_desc
=
filter_desc
.
descriptor
<
T
>
(
cudnnFilterDescriptor_t
cudnn_filter_desc
=
filter_desc
.
descriptor
<
T
>
(
layout
,
framework
::
vectorize2int
(
filter
->
dims
()));
layout
,
framework
::
vectorize2int
(
filter
->
dims
()));
...
...
paddle/platform/cudnn_helper.h
浏览文件 @
74912c7d
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
...
@@ -63,9 +60,10 @@ inline const char* cudnnGetErrorString(cudnnStatus_t status) {
...
@@ -63,9 +60,10 @@ inline const char* cudnnGetErrorString(cudnnStatus_t status) {
} \
} \
} while (false)
} while (false)
enum
class
DataLayout
{
enum
class
DataLayout
{
// Not use
kNHWC
,
kNHWC
,
kNCHW
,
kNCHW
,
kNCDHW
,
kNCHW_VECT_C
,
kNCHW_VECT_C
,
};
};
...
@@ -107,12 +105,15 @@ class CudnnDataType<double> {
...
@@ -107,12 +105,15 @@ class CudnnDataType<double> {
}
}
};
};
inline
cudnnTensorFormat_t
GetCudnnTensorFormat
(
const
DataLayout
&
order
)
{
inline
cudnnTensorFormat_t
GetCudnnTensorFormat
(
const
DataLayout
&
order
)
{
// Not use
switch
(
order
)
{
switch
(
order
)
{
case
DataLayout
::
kNHWC
:
case
DataLayout
::
kNHWC
:
return
CUDNN_TENSOR_NHWC
;
return
CUDNN_TENSOR_NHWC
;
case
DataLayout
::
kNCHW
:
case
DataLayout
::
kNCHW
:
return
CUDNN_TENSOR_NCHW
;
return
CUDNN_TENSOR_NCHW
;
case
DataLayout
::
kNCDHW
:
return
CUDNN_TENSOR_NCHW
;
// TODO(chengduoZH) : add CUDNN_TENSOR_NCDHW
default:
default:
PADDLE_THROW
(
"Unknown cudnn equivalent for order"
);
PADDLE_THROW
(
"Unknown cudnn equivalent for order"
);
}
}
...
@@ -139,7 +140,7 @@ class ScopedTensorDescriptor {
...
@@ -139,7 +140,7 @@ class ScopedTensorDescriptor {
strides
[
i
]
=
dims
[
i
+
1
]
*
strides
[
i
+
1
];
strides
[
i
]
=
dims
[
i
+
1
]
*
strides
[
i
+
1
];
}
}
// Update tensor descriptor dims setting if groups > 1
// Update tensor descriptor dims setting if groups > 1
// FIXME(typhoonzero): Assume using NCHW order
// FIXME(typhoonzero): Assume using NCHW or
NCDHW or
der
std
::
vector
<
int
>
dims_with_group
(
dims
.
begin
(),
dims
.
end
());
// copy
std
::
vector
<
int
>
dims_with_group
(
dims
.
begin
(),
dims
.
end
());
// copy
if
(
groups
>
1
)
{
if
(
groups
>
1
)
{
dims_with_group
[
1
]
=
dims_with_group
[
1
]
/
groups
;
dims_with_group
[
1
]
=
dims_with_group
[
1
]
/
groups
;
...
...
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