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
Paddle
提交
8de336f9
P
Paddle
项目概览
PaddlePaddle
/
Paddle
接近 1 年 前同步成功
通知
2292
Star
20931
Fork
5422
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1423
列表
看板
标记
里程碑
合并请求
543
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
1,423
Issue
1,423
列表
看板
标记
里程碑
合并请求
543
合并请求
543
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 >>
未验证
提交
8de336f9
编写于
12月 06, 2022
作者:
H
houj04
提交者:
GitHub
12月 06, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
[XPU] add tile_grad op (#48720)
上级
8fb829ba
变更
4
隐藏空白更改
内联
并排
Showing
4 changed file
with
111 addition
and
2 deletion
+111
-2
paddle/fluid/platform/device/xpu/xpu2_op_list.h
paddle/fluid/platform/device/xpu/xpu2_op_list.h
+1
-0
paddle/phi/kernels/impl/tile_grad_kernel_impl.h
paddle/phi/kernels/impl/tile_grad_kernel_impl.h
+2
-2
paddle/phi/kernels/xpu/tile_grad_kernel.cc
paddle/phi/kernels/xpu/tile_grad_kernel.cc
+99
-0
python/paddle/fluid/tests/unittests/xpu/test_tile_op_xpu.py
python/paddle/fluid/tests/unittests/xpu/test_tile_op_xpu.py
+9
-0
未找到文件。
paddle/fluid/platform/device/xpu/xpu2_op_list.h
浏览文件 @
8de336f9
...
...
@@ -663,6 +663,7 @@ XPUOpMap& get_kl2_ops() {
pOpKernelType
(
vartype
::
INT64
,
XPUPlace
()),
pOpKernelType
(
vartype
::
BOOL
,
XPUPlace
()),
pOpKernelType
(
vartype
::
FP32
,
XPUPlace
())})},
{
"tile_grad"
,
XPUKernelSet
({
pOpKernelType
(
vartype
::
FP32
,
XPUPlace
())})},
{
"transpose2_grad"
,
XPUKernelSet
({
pOpKernelType
(
vartype
::
FP32
,
XPUPlace
()),
pOpKernelType
(
vartype
::
FP16
,
XPUPlace
())})},
...
...
paddle/phi/kernels/impl/tile_grad_kernel_impl.h
浏览文件 @
8de336f9
...
...
@@ -97,8 +97,8 @@ void TileGradKernel(const Context& dev_ctx,
PADDLE_ENFORCE_GE
(
dims
,
1
,
errors
::
InvalidArgument
(
"Th rank of the input 'Out@GRAD' for tile_grad op "
"
must be greater than or equal to 1, but "
"Th
e
rank of the input 'Out@GRAD' for tile_grad op "
"must be greater than or equal to 1, but "
"the value received is %d."
,
dims
));
PADDLE_ENFORCE_LE
(
dims
,
...
...
paddle/phi/kernels/xpu/tile_grad_kernel.cc
0 → 100644
浏览文件 @
8de336f9
// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "paddle/phi/kernels/tile_grad_kernel.h"
#include "paddle/phi/backends/xpu/enforce_xpu.h"
#include "paddle/phi/core/kernel_registry.h"
namespace
phi
{
template
<
typename
T
,
typename
Context
>
void
TileGradKernel
(
const
Context
&
dev_ctx
,
const
DenseTensor
&
x
,
const
DenseTensor
&
out_grad
,
const
IntArray
&
repeat_times
,
DenseTensor
*
x_grad
)
{
auto
x_dims
=
x
.
dims
();
auto
vec_x_dims
=
phi
::
vectorize
<
int
>
(
x_dims
);
auto
repeat_times_data
=
repeat_times
.
GetData
();
if
(
repeat_times_data
.
size
()
<
vec_x_dims
.
size
())
{
int
diff
=
vec_x_dims
.
size
()
-
repeat_times_data
.
size
();
repeat_times_data
.
insert
(
repeat_times_data
.
begin
(),
diff
,
1
);
}
else
{
int
diff
=
repeat_times_data
.
size
()
-
vec_x_dims
.
size
();
vec_x_dims
.
insert
(
vec_x_dims
.
begin
(),
diff
,
1
);
}
// 1. reshape_dims_vec is the broadcast parameter.
// 2. reduce_dims_vec is the dimension parameter to compute gradients. For
// each dimension expanded, the gradients should be summed to original
// size.
std
::
vector
<
int
>
reshape_dims_vec
;
std
::
vector
<
int
>
reduce_dims_vec
;
for
(
size_t
i
=
0
;
i
<
repeat_times_data
.
size
();
++
i
)
{
reduce_dims_vec
.
push_back
(
reshape_dims_vec
.
size
());
reshape_dims_vec
.
push_back
(
repeat_times_data
[
i
]);
reshape_dims_vec
.
push_back
(
vec_x_dims
[
i
]);
}
dev_ctx
.
template
Alloc
<
T
>(
x_grad
);
int
dims
=
reduce_dims_vec
.
size
();
bool
just_copy
=
true
;
for
(
size_t
i
=
0
;
i
<
repeat_times_data
.
size
();
i
++
)
{
if
(
repeat_times_data
[
i
]
!=
1
)
{
just_copy
=
false
;
break
;
}
}
// no need reduce, just copy
if
(
just_copy
)
{
phi
::
Copy
(
dev_ctx
,
out_grad
,
dev_ctx
.
GetPlace
(),
false
,
x_grad
);
// TensorCopy may change the dims of dx
x_grad
->
Resize
(
x_dims
);
}
else
{
PADDLE_ENFORCE_GE
(
dims
,
1
,
errors
::
InvalidArgument
(
"The rank of the input 'Out@GRAD' for tile_grad op "
"must be greater than or equal to 1, but "
"the value received is %d."
,
dims
));
PADDLE_ENFORCE_LE
(
dims
,
MAX_RANK_SUPPORTED
,
errors
::
InvalidArgument
(
"The rank of the input 'Out@GRAD' for tile_grad op "
"must be less than or equal "
"to %d, but the value received is %d."
,
MAX_RANK_SUPPORTED
,
dims
));
using
XPUType
=
typename
XPUTypeTrait
<
T
>::
Type
;
// int reduce_sum(Context* ctx, const T* x, T* y, const std::vector<int>&
// xshape, const std::vector<int>& rdims)
const
auto
*
out_data
=
out_grad
.
data
<
XPUType
>
();
auto
*
x_grad_data
=
x_grad
->
data
<
XPUType
>
();
int
r
=
xpu
::
reduce_sum
<
XPUType
>
(
dev_ctx
.
x_context
(),
out_data
,
x_grad_data
,
reshape_dims_vec
,
reduce_dims_vec
);
PADDLE_ENFORCE_XDNN_SUCCESS
(
r
,
"reduce_sum"
);
}
}
}
// namespace phi
PD_REGISTER_KERNEL
(
tile_grad
,
XPU
,
ALL_LAYOUT
,
phi
::
TileGradKernel
,
float
)
{}
python/paddle/fluid/tests/unittests/xpu/test_tile_op_xpu.py
浏览文件 @
8de336f9
...
...
@@ -59,6 +59,9 @@ class XPUTestTileOpRank1(XPUOpTestWrapper):
def
test_check_output
(
self
):
self
.
check_output_with_place
(
self
.
place
)
def
test_check_grad
(
self
):
self
.
check_grad_with_place
(
self
.
place
,
[
'X'
],
'Out'
)
# with dimension expanding
class
TestTileOpRank2Expanding
(
TestTileOpRank1
):
def
init_data
(
self
):
...
...
@@ -126,6 +129,9 @@ class XPUTestTileOpRank1_tensor_attr(XPUOpTestWrapper):
def
test_check_output
(
self
):
self
.
check_output_with_place
(
self
.
place
)
def
test_check_grad
(
self
):
self
.
check_grad_with_place
(
self
.
place
,
[
'X'
],
'Out'
)
class
TestTileOpRank2_Corner_tensor_attr
(
TestTileOpRank1_tensor_attr
):
def
init_data
(
self
):
self
.
ori_shape
=
[
12
,
14
]
...
...
@@ -168,6 +174,9 @@ class XPUTestTileOpRank1_tensor(XPUOpTestWrapper):
def
test_check_output
(
self
):
self
.
check_output_with_place
(
self
.
place
)
def
test_check_grad
(
self
):
self
.
check_grad_with_place
(
self
.
place
,
[
'X'
],
'Out'
)
class
TestTileOpRank2_tensor
(
TestTileOpRank1_tensor
):
def
init_data
(
self
):
self
.
ori_shape
=
[
12
,
14
]
...
...
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