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
飞不过沧海
Paddle
提交
94d8f392
P
Paddle
项目概览
飞不过沧海
/
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
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
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
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 >>
未验证
提交
94d8f392
编写于
2月 25, 2022
作者:
C
Chen Weihang
提交者:
GitHub
2月 25, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
move for_range into phi (#39931)
上级
687902fc
变更
14
隐藏空白更改
内联
并排
Showing
14 changed file
with
160 addition
and
152 deletion
+160
-152
paddle/fluid/platform/for_range.h
paddle/fluid/platform/for_range.h
+3
-124
paddle/phi/kernels/cpu/abs_kernel.cc
paddle/phi/kernels/cpu/abs_kernel.cc
+2
-2
paddle/phi/kernels/funcs/diagonal.h
paddle/phi/kernels/funcs/diagonal.h
+2
-2
paddle/phi/kernels/funcs/elementwise_base.h
paddle/phi/kernels/funcs/elementwise_base.h
+2
-2
paddle/phi/kernels/funcs/for_range.h
paddle/phi/kernels/funcs/for_range.h
+129
-0
paddle/phi/kernels/gpu/poisson_kernel.cu
paddle/phi/kernels/gpu/poisson_kernel.cu
+2
-2
paddle/phi/kernels/impl/abs_grad_kernel_impl.h
paddle/phi/kernels/impl/abs_grad_kernel_impl.h
+3
-3
paddle/phi/kernels/impl/atan2_grad_kernel_impl.h
paddle/phi/kernels/impl/atan2_grad_kernel_impl.h
+2
-2
paddle/phi/kernels/impl/atan2_kernel_impl.h
paddle/phi/kernels/impl/atan2_kernel_impl.h
+2
-2
paddle/phi/kernels/impl/complex_grad_kernel_impl.h
paddle/phi/kernels/impl/complex_grad_kernel_impl.h
+3
-3
paddle/phi/kernels/impl/complex_kernel_impl.h
paddle/phi/kernels/impl/complex_kernel_impl.h
+4
-4
paddle/phi/kernels/impl/digamma_grad_kernel_impl.h
paddle/phi/kernels/impl/digamma_grad_kernel_impl.h
+2
-2
paddle/phi/kernels/impl/digamma_kernel_impl.h
paddle/phi/kernels/impl/digamma_kernel_impl.h
+2
-2
paddle/phi/kernels/impl/trace_grad_kernel_impl.h
paddle/phi/kernels/impl/trace_grad_kernel_impl.h
+2
-2
未找到文件。
paddle/fluid/platform/for_range.h
浏览文件 @
94d8f392
...
...
@@ -13,136 +13,15 @@ See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include "paddle/fluid/platform/device/gpu/gpu_launch_config.h"
#include "paddle/fluid/platform/device_context.h"
#include "paddle/phi/backends/cpu/cpu_context.h"
#include "paddle/phi/backends/gpu/gpu_context.h"
#include "paddle/phi/kernels/funcs/for_range.h"
namespace
paddle
{
namespace
platform
{
template
<
typename
DeviceContext
>
struct
ForRange
{
ForRange
(
const
DeviceContext
&
dev_ctx
,
size_t
limit
);
template
<
typename
Function
>
void
operator
()(
Function
func
)
const
;
};
// NOTE: After the pten kernel is migrated, it needs to be deleted.
template
<
>
struct
ForRange
<
CPUDeviceContext
>
{
ForRange
(
const
CPUDeviceContext
&
dev_ctx
,
size_t
limit
)
:
limit_
(
limit
)
{}
template
<
typename
Function
>
void
operator
()(
Function
func
)
const
{
for
(
size_t
i
=
0
;
i
<
limit_
;
++
i
)
{
func
(
i
);
}
}
size_t
limit_
;
};
template
<
>
struct
ForRange
<
phi
::
CPUContext
>
{
ForRange
(
const
phi
::
CPUContext
&
dev_ctx
,
size_t
limit
)
:
limit_
(
limit
)
{}
template
<
typename
Function
>
void
operator
()(
Function
func
)
const
{
for
(
size_t
i
=
0
;
i
<
limit_
;
++
i
)
{
func
(
i
);
}
}
size_t
limit_
;
};
#if defined(__NVCC__) || defined(__HIPCC__)
template
<
typename
Function
>
__global__
static
void
ForRangeElemwiseOpGridIsOne
(
Function
func
)
{
size_t
idx
=
static_cast
<
size_t
>
(
threadIdx
.
x
);
func
(
idx
);
}
template
<
typename
Function
>
__global__
static
void
ForRangeElemwiseOp
(
Function
func
,
size_t
limit
)
{
size_t
idx
=
static_cast
<
size_t
>
(
blockIdx
.
x
*
blockDim
.
x
+
threadIdx
.
x
);
if
(
idx
<
limit
)
{
func
(
idx
);
}
}
// NOTE: After the pten kernel is migrated, it needs to be deleted.
template
<
>
struct
ForRange
<
CUDADeviceContext
>
{
ForRange
(
const
CUDADeviceContext
&
dev_ctx
,
size_t
limit
)
:
dev_ctx_
(
dev_ctx
),
limit_
(
static_cast
<
size_t
>
(
limit
))
{}
template
<
typename
Function
>
inline
void
operator
()(
Function
func
)
const
{
#ifdef __HIPCC__
// HIP will throw core dump when threads > 256
constexpr
int
num_threads
=
256
;
#elif WITH_NV_JETSON
// JETSON_NANO will throw core dump when threads > 128
int
num_thread
=
256
;
platform
::
ChangeThreadNum
(
dev_ctx_
,
&
num_thread
,
128
);
const
int
num_threads
=
num_thread
;
#else
constexpr
int
num_threads
=
1024
;
#endif
size_t
block_size
=
limit_
<=
num_threads
?
limit_
:
num_threads
;
size_t
grid_size
=
(
limit_
+
num_threads
-
1
)
/
num_threads
;
if
(
grid_size
==
1
)
{
ForRangeElemwiseOpGridIsOne
<<<
1
,
block_size
,
0
,
dev_ctx_
.
stream
()
>>>
(
func
);
}
else
{
ForRangeElemwiseOp
<<<
grid_size
,
block_size
,
0
,
dev_ctx_
.
stream
()
>>>
(
func
,
limit_
);
}
}
const
CUDADeviceContext
&
dev_ctx_
;
size_t
limit_
;
};
template
<
>
struct
ForRange
<
phi
::
GPUContext
>
{
ForRange
(
const
phi
::
GPUContext
&
dev_ctx
,
size_t
limit
)
:
dev_ctx_
(
dev_ctx
),
limit_
(
static_cast
<
size_t
>
(
limit
))
{}
template
<
typename
Function
>
inline
void
operator
()(
Function
func
)
const
{
#ifdef __HIPCC__
// HIP will throw core dump when threads > 256
constexpr
int
num_threads
=
256
;
#elif WITH_NV_JETSON
// JETSON_NANO will throw core dump when threads > 128
int
num_thread
=
256
;
platform
::
ChangeThreadNum
(
dev_ctx_
,
&
num_thread
,
128
);
const
int
num_threads
=
num_thread
;
#else
constexpr
int
num_threads
=
1024
;
#endif
size_t
block_size
=
limit_
<=
num_threads
?
limit_
:
num_threads
;
size_t
grid_size
=
(
limit_
+
num_threads
-
1
)
/
num_threads
;
if
(
grid_size
==
1
)
{
ForRangeElemwiseOpGridIsOne
<<<
1
,
block_size
,
0
,
dev_ctx_
.
stream
()
>>>
(
func
);
}
else
{
ForRangeElemwiseOp
<<<
grid_size
,
block_size
,
0
,
dev_ctx_
.
stream
()
>>>
(
func
,
limit_
);
}
}
const
phi
::
GPUContext
&
dev_ctx_
;
size_t
limit_
;
};
#endif
using
ForRange
=
phi
::
funcs
::
ForRange
<
DeviceContext
>
;
}
// namespace platform
}
// namespace paddle
paddle/phi/kernels/cpu/abs_kernel.cc
浏览文件 @
94d8f392
...
...
@@ -13,11 +13,11 @@
// limitations under the License.
#include "paddle/phi/kernels/abs_kernel.h"
#include "paddle/fluid/platform/for_range.h"
#include "paddle/phi/backends/cpu/cpu_context.h"
#include "paddle/phi/common/complex.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/funcs/complex_functors.h"
#include "paddle/phi/kernels/funcs/for_range.h"
namespace
phi
{
...
...
@@ -29,7 +29,7 @@ void AbsKernel(const Context& ctx, const DenseTensor& x, DenseTensor* out) {
out
,
size_t
(
x
.
numel
()
*
sizeof
(
phi
::
funcs
::
Real
<
T
>
)));
auto
*
out_data
=
out
->
data
<
phi
::
funcs
::
Real
<
T
>>
();
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
phi
::
funcs
::
AbsFunctor
<
T
>
functor
(
x_data
,
out_data
,
numel
);
for_range
(
functor
);
}
...
...
paddle/phi/kernels/funcs/diagonal.h
浏览文件 @
94d8f392
...
...
@@ -22,8 +22,8 @@
#include <algorithm>
#include "paddle/fluid/platform/for_range.h"
#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/kernels/funcs/for_range.h"
namespace
phi
{
namespace
funcs
{
...
...
@@ -118,7 +118,7 @@ DenseTensor Diagonal(const DeviceContext& context,
#endif
// auto& dev_ctx = context.template device_context<DeviceContext>();
p
addle
::
platform
::
ForRange
<
DeviceContext
>
for_range
(
context
,
diag
.
numel
());
p
hi
::
funcs
::
ForRange
<
DeviceContext
>
for_range
(
context
,
diag
.
numel
());
DiagonalFunctor
<
T
>
functor
(
input_data
,
diag_arr
,
ret_arr
,
pos
,
dim_size
,
diag_data
);
for_range
(
functor
);
...
...
paddle/phi/kernels/funcs/elementwise_base.h
浏览文件 @
94d8f392
...
...
@@ -14,11 +14,11 @@ limitations under the License. */
#pragma once
#include "paddle/fluid/platform/for_range.h"
#include "paddle/fluid/platform/transform.h"
#include "paddle/phi/backends/all_context.h"
#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/kernels/empty_kernel.h"
#include "paddle/phi/kernels/funcs/for_range.h"
#include "paddle/phi/kernels/funcs/math_function.h"
#if defined(__NVCC__) || defined(__HIPCC__) || defined(__xpu__)
...
...
@@ -418,7 +418,7 @@ void ElemwiseGradComputeNoBroadcast(const DeviceContext &dev_ctx,
DX_OP
dx_op
,
DY_OP
dy_op
)
{
size_t
N
=
static_cast
<
size_t
>
(
phi
::
product
(
x_dim
));
p
addle
::
platform
::
ForRange
<
DeviceContext
>
for_range
(
dev_ctx
,
N
);
p
hi
::
funcs
::
ForRange
<
DeviceContext
>
for_range
(
dev_ctx
,
N
);
for_range
(
ElemwiseGradNoBroadcast
<
T
,
DX_OP
,
DY_OP
,
Tout
>
{
x
.
data
<
T
>
(),
y
.
data
<
T
>
(),
...
...
paddle/phi/kernels/funcs/for_range.h
0 → 100644
浏览文件 @
94d8f392
/* Copyright (c) 2016 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. */
#pragma once
#include "paddle/phi/backends/all_context.h"
#include "paddle/phi/backends/gpu/gpu_launch_config.h"
namespace
phi
{
namespace
funcs
{
template
<
typename
Context
>
struct
ForRange
{
ForRange
(
const
Context
&
dev_ctx
,
size_t
limit
);
template
<
typename
Function
>
void
operator
()(
Function
func
)
const
;
};
template
<
>
struct
ForRange
<
phi
::
CPUContext
>
{
ForRange
(
const
phi
::
CPUContext
&
dev_ctx
,
size_t
limit
)
:
limit_
(
limit
)
{}
template
<
typename
Function
>
void
operator
()(
Function
func
)
const
{
for
(
size_t
i
=
0
;
i
<
limit_
;
++
i
)
{
func
(
i
);
}
}
size_t
limit_
;
};
// NOTE: After the pten kernel is migrated, it needs to be deleted.
template
<
>
struct
ForRange
<
paddle
::
platform
::
CPUDeviceContext
>
{
ForRange
(
const
paddle
::
platform
::
CPUDeviceContext
&
dev_ctx
,
size_t
limit
)
:
dev_ctx_
(
dev_ctx
),
limit_
(
limit
)
{}
template
<
typename
Function
>
void
operator
()(
Function
func
)
const
{
phi
::
funcs
::
ForRange
<
phi
::
CPUContext
>
for_range
(
dev_ctx_
,
limit_
);
for_range
(
func
);
}
const
paddle
::
platform
::
CPUDeviceContext
&
dev_ctx_
;
size_t
limit_
;
};
#if defined(__NVCC__) || defined(__HIPCC__)
template
<
typename
Function
>
__global__
static
void
ForRangeElemwiseOpGridIsOne
(
Function
func
)
{
size_t
idx
=
static_cast
<
size_t
>
(
threadIdx
.
x
);
func
(
idx
);
}
template
<
typename
Function
>
__global__
static
void
ForRangeElemwiseOp
(
Function
func
,
size_t
limit
)
{
size_t
idx
=
static_cast
<
size_t
>
(
blockIdx
.
x
*
blockDim
.
x
+
threadIdx
.
x
);
if
(
idx
<
limit
)
{
func
(
idx
);
}
}
template
<
>
struct
ForRange
<
phi
::
GPUContext
>
{
ForRange
(
const
phi
::
GPUContext
&
dev_ctx
,
size_t
limit
)
:
dev_ctx_
(
dev_ctx
),
limit_
(
limit
)
{}
template
<
typename
Function
>
inline
void
operator
()(
Function
func
)
const
{
#ifdef __HIPCC__
// HIP will throw core dump when threads > 256
constexpr
int
num_threads
=
256
;
#elif WITH_NV_JETSON
// JETSON_NANO will throw core dump when threads > 128
int
num_thread
=
256
;
backends
::
gpu
::
ChangeThreadNum
(
dev_ctx_
,
&
num_thread
,
128
);
const
int
num_threads
=
num_thread
;
#else
constexpr
int
num_threads
=
1024
;
#endif
size_t
block_size
=
limit_
<=
num_threads
?
limit_
:
num_threads
;
size_t
grid_size
=
(
limit_
+
num_threads
-
1
)
/
num_threads
;
if
(
grid_size
==
1
)
{
ForRangeElemwiseOpGridIsOne
<<<
1
,
block_size
,
0
,
dev_ctx_
.
stream
()
>>>
(
func
);
}
else
{
ForRangeElemwiseOp
<<<
grid_size
,
block_size
,
0
,
dev_ctx_
.
stream
()
>>>
(
func
,
limit_
);
}
}
const
phi
::
GPUContext
&
dev_ctx_
;
size_t
limit_
;
};
// NOTE: After the pten kernel is migrated, it needs to be deleted.
template
<
>
struct
ForRange
<
paddle
::
platform
::
CUDADeviceContext
>
{
ForRange
(
const
paddle
::
platform
::
CUDADeviceContext
&
dev_ctx
,
size_t
limit
)
:
dev_ctx_
(
dev_ctx
),
limit_
(
limit
)
{}
template
<
typename
Function
>
inline
void
operator
()(
Function
func
)
const
{
phi
::
funcs
::
ForRange
<
phi
::
GPUContext
>
for_range
(
dev_ctx_
,
limit_
);
for_range
(
func
);
}
const
paddle
::
platform
::
CUDADeviceContext
&
dev_ctx_
;
size_t
limit_
;
};
#endif
}
// namespace funcs
}
// namespace phi
paddle/phi/kernels/gpu/poisson_kernel.cu
浏览文件 @
94d8f392
...
...
@@ -19,9 +19,9 @@ limitations under the License. */
#include <hiprand_kernel.h>
#endif
#include "paddle/fluid/platform/for_range.h"
#include "paddle/phi/backends/gpu/gpu_context.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/funcs/for_range.h"
#include "paddle/phi/kernels/poisson_kernel.h"
namespace
phi
{
...
...
@@ -65,7 +65,7 @@ void PoissonKernel(const Context& ctx, const DenseTensor& x, DenseTensor* out) {
uint64_t
seed
=
seed_offset
.
first
;
uint64_t
offset
=
seed_offset
.
second
;
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
ctx
,
size
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
ctx
,
size
);
PoissonCudaFunctor
<
T
>
functor
(
x_data
,
out_data
,
seed
,
offset
);
for_range
(
functor
);
...
...
paddle/phi/kernels/impl/abs_grad_kernel_impl.h
浏览文件 @
94d8f392
...
...
@@ -14,10 +14,10 @@
#pragma once
#include "paddle/fluid/platform/for_range.h"
#include "paddle/phi/kernels/abs_grad_kernel.h"
#include "paddle/phi/kernels/funcs/complex_functors.h"
#include "paddle/phi/kernels/funcs/elementwise_base.h"
#include "paddle/phi/kernels/funcs/for_range.h"
namespace
phi
{
...
...
@@ -53,7 +53,7 @@ void AbsGradKernel(const Context& ctx,
ctx
.
template
Alloc
<
T
>(
dx
,
static_cast
<
size_t
>
(
numel
*
sizeof
(
T
)));
auto
*
dx_data
=
dx
->
data
<
T
>
();
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
phi
::
funcs
::
AbsGradFunctor
<
T
>
functor
(
dout_data
,
x_data
,
dx_data
,
numel
);
for_range
(
functor
);
}
...
...
@@ -70,7 +70,7 @@ void AbsDoubleGradKernel(const Context& ctx,
ctx
.
template
Alloc
<
T
>(
ddout
,
static_cast
<
size_t
>
(
numel
*
sizeof
(
T
)));
auto
*
ddout_data
=
ddout
->
data
<
T
>
();
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
phi
::
funcs
::
AbsGradGradFunctor
<
T
>
functor
(
ddx_data
,
x_data
,
ddout_data
,
numel
);
for_range
(
functor
);
...
...
paddle/phi/kernels/impl/atan2_grad_kernel_impl.h
浏览文件 @
94d8f392
...
...
@@ -14,9 +14,9 @@
#pragma once
#include "paddle/fluid/platform/for_range.h"
#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/kernels/atan2_grad_kernel.h"
#include "paddle/phi/kernels/funcs/for_range.h"
namespace
phi
{
...
...
@@ -85,7 +85,7 @@ void Atan2GradKernel(const Context& ctx,
auto
*
y_grad_data
=
ctx
.
template
Alloc
<
T
>(
y_grad
,
size_t
(
y
.
numel
()
*
sizeof
(
T
)));
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
phi
::
Atan2GradFunctor
<
T
>
functor
(
x_data
,
y_data
,
out_grad_data
,
x_grad_data
,
y_grad_data
,
numel
);
for_range
(
functor
);
...
...
paddle/phi/kernels/impl/atan2_kernel_impl.h
浏览文件 @
94d8f392
...
...
@@ -14,9 +14,9 @@
#pragma once
#include "paddle/fluid/platform/for_range.h"
#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/kernels/atan2_kernel.h"
#include "paddle/phi/kernels/funcs/for_range.h"
namespace
phi
{
template
<
typename
T
>
...
...
@@ -80,7 +80,7 @@ void Atan2Kernel(const Context& ctx,
auto
*
out_data
=
ctx
.
template
Alloc
<
typename
Atan2Out
<
T
>
::
type
>
(
out
,
size_t
(
x
.
numel
()
*
sizeof
(
typename
Atan2Out
<
T
>::
type
)));
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
phi
::
Atan2Functor
<
T
>
functor
(
x_data
,
y_data
,
out_data
,
numel
);
for_range
(
functor
);
}
...
...
paddle/phi/kernels/impl/complex_grad_kernel_impl.h
浏览文件 @
94d8f392
...
...
@@ -14,8 +14,8 @@
#pragma once
#include "paddle/fluid/platform/for_range.h"
#include "paddle/phi/kernels/funcs/complex_functors.h"
#include "paddle/phi/kernels/funcs/for_range.h"
namespace
phi
{
...
...
@@ -28,7 +28,7 @@ void RealGradKernel(const Context& dev_ctx,
auto
*
dx_data
=
dev_ctx
.
template
Alloc
<
T
>(
dx
,
static_cast
<
size_t
>
(
numel
*
sizeof
(
T
)));
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
dev_ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
dev_ctx
,
numel
);
phi
::
funcs
::
RealToComplexFunctor
<
T
>
functor
(
dout_data
,
dx_data
,
numel
);
for_range
(
functor
);
}
...
...
@@ -42,7 +42,7 @@ void ImagGradKernel(const Context& dev_ctx,
auto
*
dx_data
=
dev_ctx
.
template
Alloc
<
T
>(
dx
,
static_cast
<
size_t
>
(
numel
*
sizeof
(
T
)));
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
dev_ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
dev_ctx
,
numel
);
phi
::
funcs
::
ImagToComplexFunctor
<
T
>
functor
(
dout_data
,
dx_data
,
numel
);
for_range
(
functor
);
}
...
...
paddle/phi/kernels/impl/complex_kernel_impl.h
浏览文件 @
94d8f392
...
...
@@ -15,8 +15,8 @@
#pragma once
// See Note [ Why still include the fluid headers? ]
#include "paddle/fluid/platform/for_range.h"
#include "paddle/phi/kernels/funcs/complex_functors.h"
#include "paddle/phi/kernels/funcs/for_range.h"
namespace
phi
{
...
...
@@ -28,7 +28,7 @@ void ConjKernel(const Context& dev_ctx,
auto
*
x_data
=
x
.
data
<
T
>
();
auto
*
out_data
=
dev_ctx
.
template
Alloc
<
T
>(
out
);
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
dev_ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
dev_ctx
,
numel
);
phi
::
funcs
::
ConjFunctor
<
T
>
functor
(
x_data
,
numel
,
out_data
);
for_range
(
functor
);
}
...
...
@@ -42,7 +42,7 @@ void RealKernel(const Context& dev_ctx,
auto
*
out_data
=
dev_ctx
.
template
Alloc
<
phi
::
funcs
::
Real
<
T
>
>
(
out
,
static_cast
<
size_t
>
(
numel
*
sizeof
(
phi
::
funcs
::
Real
<
T
>
)));
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
dev_ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
dev_ctx
,
numel
);
phi
::
funcs
::
RealFunctor
<
T
>
functor
(
x_data
,
out_data
,
numel
);
for_range
(
functor
);
}
...
...
@@ -56,7 +56,7 @@ void ImagKernel(const Context& dev_ctx,
auto
*
out_data
=
dev_ctx
.
template
Alloc
<
phi
::
funcs
::
Real
<
T
>
>
(
out
,
static_cast
<
size_t
>
(
numel
*
sizeof
(
phi
::
funcs
::
Real
<
T
>
)));
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
dev_ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
dev_ctx
,
numel
);
phi
::
funcs
::
ImagFunctor
<
T
>
functor
(
x_data
,
out_data
,
numel
);
for_range
(
functor
);
}
...
...
paddle/phi/kernels/impl/digamma_grad_kernel_impl.h
浏览文件 @
94d8f392
...
...
@@ -15,8 +15,8 @@
#pragma once
#include <unsupported/Eigen/SpecialFunctions>
#include "paddle/fluid/platform/for_range.h"
#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/kernels/funcs/for_range.h"
namespace
phi
{
...
...
@@ -47,7 +47,7 @@ void DigammaGradKernel(const Context& ctx,
auto
*
x_data
=
x
.
data
<
T
>
();
auto
*
dx_data
=
x_grad
->
data
<
T
>
();
auto
numel
=
out_grad
.
numel
();
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
DigammaGradFunctor
<
T
>
functor
(
dout_data
,
x_data
,
dx_data
,
numel
);
for_range
(
functor
);
}
...
...
paddle/phi/kernels/impl/digamma_kernel_impl.h
浏览文件 @
94d8f392
...
...
@@ -15,8 +15,8 @@
#pragma once
#include <unsupported/Eigen/SpecialFunctions>
#include "paddle/fluid/platform/for_range.h"
#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/kernels/funcs/for_range.h"
namespace
phi
{
...
...
@@ -41,7 +41,7 @@ void DigammaKernel(const Context& ctx, const DenseTensor& x, DenseTensor* out) {
auto
*
x_data
=
x
.
data
<
T
>
();
auto
*
out_data
=
out
->
data
<
T
>
();
auto
numel
=
x
.
numel
();
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
ctx
,
numel
);
DigammaFunctor
<
T
>
functor
(
x_data
,
out_data
,
numel
);
for_range
(
functor
);
}
...
...
paddle/phi/kernels/impl/trace_grad_kernel_impl.h
浏览文件 @
94d8f392
...
...
@@ -21,7 +21,7 @@
#include <algorithm>
#include "paddle/
fluid/platform
/for_range.h"
#include "paddle/
phi/kernels/funcs
/for_range.h"
#include "paddle/phi/kernels/funcs/math_function.h"
namespace
phi
{
...
...
@@ -130,7 +130,7 @@ void TraceGradKernel(const Context& ctx,
const
auto
*
input_arr
=
input_stride
.
Get
();
#endif
p
addle
::
platform
::
ForRange
<
Context
>
for_range
(
ctx
,
in_grad
->
numel
());
p
hi
::
funcs
::
ForRange
<
Context
>
for_range
(
ctx
,
in_grad
->
numel
());
TraceGradFunctor
<
T
>
functor
(
out_data
,
output_arr
,
input_arr
,
...
...
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