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
37ca3b4c
P
Paddle
项目概览
PaddlePaddle
/
Paddle
大约 1 年 前同步成功
通知
2297
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
37ca3b4c
编写于
4月 18, 2023
作者:
H
huangjiyi
提交者:
GitHub
4月 18, 2023
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
register fluid kerenls to phi [part 6.4] (#52881)
* update * revert lookup_table_op
上级
4d5a3ad6
变更
13
隐藏空白更改
内联
并排
Showing
13 changed file
with
92 addition
and
71 deletion
+92
-71
paddle/fluid/operators/custom_device_common_op_registry.cc
paddle/fluid/operators/custom_device_common_op_registry.cc
+5
-5
paddle/fluid/operators/load_combine_op.cc
paddle/fluid/operators/load_combine_op.cc
+11
-8
paddle/fluid/operators/load_combine_op.cu
paddle/fluid/operators/load_combine_op.cu
+9
-7
paddle/fluid/operators/load_combine_op.h
paddle/fluid/operators/load_combine_op.h
+1
-1
paddle/fluid/operators/load_combine_op_xpu.cc
paddle/fluid/operators/load_combine_op_xpu.cc
+7
-7
paddle/fluid/operators/lod_reset_op.cc
paddle/fluid/operators/lod_reset_op.cc
+26
-23
paddle/fluid/operators/lod_reset_op.cu
paddle/fluid/operators/lod_reset_op.cu
+16
-10
paddle/fluid/operators/lod_reset_op.h
paddle/fluid/operators/lod_reset_op.h
+2
-2
paddle/fluid/operators/lookup_table_dequant_op.cc
paddle/fluid/operators/lookup_table_dequant_op.cc
+6
-2
paddle/fluid/operators/lookup_table_dequant_op.h
paddle/fluid/operators/lookup_table_dequant_op.h
+1
-1
paddle/fluid/operators/marker_op.cc
paddle/fluid/operators/marker_op.cc
+3
-2
paddle/fluid/operators/marker_op.cu
paddle/fluid/operators/marker_op.cu
+3
-2
paddle/fluid/operators/save_load_combine_op_test.cc
paddle/fluid/operators/save_load_combine_op_test.cc
+2
-1
未找到文件。
paddle/fluid/operators/custom_device_common_op_registry.cc
浏览文件 @
37ca3b4c
...
...
@@ -521,15 +521,15 @@ void RegisterCustomDeviceCommonKernel(const std::string& dev_type) {
load_combine
,
device_type
,
paddle
::
operators
::
LoadCombineOpKernel
<
paddle
::
platform
::
CustomDeviceContext
,
floa
t
>
,
LoadCombineOpKernel
<
float
,
paddle
::
platform
::
CustomDeviceContex
t
>
,
paddle
::
operators
::
LoadCombineOpKernel
<
paddle
::
platform
::
CustomDeviceContext
,
double
>
,
LoadCombineOpKernel
<
double
,
paddle
::
platform
::
CustomDeviceContext
>
,
paddle
::
operators
::
LoadCombineOpKernel
<
paddle
::
platform
::
CustomDeviceContext
,
in
t
>
,
LoadCombineOpKernel
<
int
,
paddle
::
platform
::
CustomDeviceContex
t
>
,
paddle
::
operators
::
LoadCombineOpKernel
<
paddle
::
platform
::
CustomDeviceContext
,
int8_
t
>
,
LoadCombineOpKernel
<
int8_t
,
paddle
::
platform
::
CustomDeviceContex
t
>
,
paddle
::
operators
::
LoadCombineOpKernel
<
paddle
::
platform
::
CustomDeviceContext
,
int64_
t
>
);
LoadCombineOpKernel
<
int64_t
,
paddle
::
platform
::
CustomDeviceContex
t
>
);
REGISTER_OP_CUSTOM_DEVICE_KERNEL
(
c_concat
,
device_type
,
...
...
paddle/fluid/operators/load_combine_op.cc
浏览文件 @
37ca3b4c
...
...
@@ -77,16 +77,19 @@ that were saved using the SaveCombine operator.
}
// namespace paddle
namespace
ops
=
paddle
::
operators
;
namespace
plat
=
paddle
::
platform
;
REGISTER_OPERATOR
(
load_combine
,
ops
::
LoadCombineOp
,
ops
::
LoadCombineOpProtoMaker
);
REGISTER_OP_CPU_KERNEL
(
load_combine
,
ops
::
LoadCombineOpKernel
<
phi
::
CPUContext
,
float
>
,
ops
::
LoadCombineOpKernel
<
phi
::
CPUContext
,
double
>
,
ops
::
LoadCombineOpKernel
<
phi
::
CPUContext
,
paddle
::
platform
::
bfloat16
>
,
ops
::
LoadCombineOpKernel
<
phi
::
CPUContext
,
int
>
,
ops
::
LoadCombineOpKernel
<
phi
::
CPUContext
,
int8_t
>
,
ops
::
LoadCombineOpKernel
<
phi
::
CPUContext
,
int64_t
>
);
PD_REGISTER_STRUCT_KERNEL
(
load_combine
,
CPU
,
ALL_LAYOUT
,
ops
::
LoadCombineOpKernel
,
float
,
double
,
plat
::
bfloat16
,
int
,
int8_t
,
int64_t
)
{}
paddle/fluid/operators/load_combine_op.cu
浏览文件 @
37ca3b4c
...
...
@@ -15,10 +15,12 @@ limitations under the License. */
#include "paddle/fluid/operators/load_combine_op.h"
namespace
ops
=
paddle
::
operators
;
REGISTER_OP_CUDA_KERNEL
(
load_combine
,
ops
::
LoadCombineOpKernel
<
phi
::
GPUContext
,
float
>
,
ops
::
LoadCombineOpKernel
<
phi
::
GPUContext
,
double
>
,
ops
::
LoadCombineOpKernel
<
phi
::
GPUContext
,
int
>
,
ops
::
LoadCombineOpKernel
<
phi
::
GPUContext
,
int8_t
>
,
ops
::
LoadCombineOpKernel
<
phi
::
GPUContext
,
int64_t
>
);
PD_REGISTER_STRUCT_KERNEL
(
load_combine
,
GPU
,
ALL_LAYOUT
,
ops
::
LoadCombineOpKernel
,
float
,
double
,
int
,
int8_t
,
int64_t
)
{}
paddle/fluid/operators/load_combine_op.h
浏览文件 @
37ca3b4c
...
...
@@ -28,7 +28,7 @@ limitations under the License. */
namespace
paddle
{
namespace
operators
{
template
<
typename
DeviceContext
,
typename
T
>
template
<
typename
T
,
typename
DeviceContext
>
class
LoadCombineOpKernel
:
public
framework
::
OpKernel
<
T
>
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
ctx
)
const
override
{
...
...
paddle/fluid/operators/load_combine_op_xpu.cc
浏览文件 @
37ca3b4c
...
...
@@ -15,11 +15,11 @@ limitations under the License. */
#include "paddle/fluid/operators/load_combine_op.h"
namespace
ops
=
paddle
::
operators
;
using
XPUCtx
=
paddle
::
platform
::
XPUDeviceContext
;
REGISTER_OP_XPU_KERNEL
(
load_combine
,
ops
::
LoadCombineOpKernel
<
paddle
::
platform
::
XPUDeviceContext
,
float
>
,
ops
::
LoadCombineOpKernel
<
paddle
::
platform
::
XPUDeviceContext
,
double
>
,
ops
::
LoadCombineOpKernel
<
paddle
::
platform
::
XPUDeviceContext
,
int
>
,
ops
::
LoadCombineOpKernel
<
paddle
::
platform
::
XPUDeviceContext
,
int8_t
>
,
ops
::
LoadCombineOpKernel
<
paddle
::
platform
::
XPUDeviceContext
,
int64_t
>
);
REGISTER_OP_XPU_KERNEL
(
load_combine
,
ops
::
LoadCombineOpKernel
<
float
,
XPUCtx
>
,
ops
::
LoadCombineOpKernel
<
double
,
XPUCtx
>
,
ops
::
LoadCombineOpKernel
<
int
,
XPUCtx
>
,
ops
::
LoadCombineOpKernel
<
int8_t
,
XPUCtx
>
,
ops
::
LoadCombineOpKernel
<
int64_t
,
XPUCtx
>
);
paddle/fluid/operators/lod_reset_op.cc
浏览文件 @
37ca3b4c
...
...
@@ -235,6 +235,7 @@ DECLARE_NO_NEED_BUFFER_VARS_INFERER(LoDResetGradNoNeedBufferVarInferer, "X");
}
// namespace paddle
namespace
ops
=
paddle
::
operators
;
namespace
plat
=
paddle
::
platform
;
REGISTER_OPERATOR
(
lod_reset
,
ops
::
LoDResetOp
,
ops
::
LoDResetOpMaker
,
...
...
@@ -247,30 +248,32 @@ REGISTER_OPERATOR(lod_reset_grad,
ops
::
LoDResetGradNoNeedBufferVarInferer
,
ops
::
LoDResetGradInplaceInferer
);
REGISTER_OP_CPU_KERNEL
(
lod_reset
,
ops
::
LoDResetKernel
<
paddle
::
platform
::
CPUPlace
,
paddle
::
platform
::
float16
>
,
ops
::
LoDResetKernel
<
paddle
::
platform
::
CPUPlace
,
float
>
,
ops
::
LoDResetKernel
<
paddle
::
platform
::
CPUPlace
,
double
>
,
ops
::
LoDResetKernel
<
paddle
::
platform
::
CPUPlace
,
int
>
,
ops
::
LoDResetKernel
<
paddle
::
platform
::
CPUPlace
,
int64_t
>
);
PD_REGISTER_STRUCT_KERNEL
(
lod_reset
,
CPU
,
ALL_LAYOUT
,
ops
::
LoDResetKernel
,
plat
::
float16
,
float
,
double
,
int
,
int64_t
)
{}
#ifdef PADDLE_WITH_XPU
REGISTER_OP_XPU_KERNEL
(
lod_reset
,
ops
::
LoDResetKernel
<
paddle
::
platform
::
XPUDeviceContext
,
paddle
::
platform
::
float16
>
,
ops
::
LoDResetKernel
<
paddle
::
platform
::
XPUDeviceContext
,
float
>
,
ops
::
LoDResetKernel
<
paddle
::
platform
::
XPUDeviceContext
,
double
>
,
ops
::
LoDResetKernel
<
paddle
::
platform
::
XPUDeviceContext
,
int
>
,
ops
::
LoDResetKernel
<
paddle
::
platform
::
XPUDeviceContext
,
int64_t
>
);
using
XPUCtx
=
paddle
::
platform
::
XPUDeviceContext
;
REGISTER_OP_XPU_KERNEL
(
lod_reset
,
ops
::
LoDResetKernel
<
paddle
::
platform
::
float16
,
XPUCtx
>
,
ops
::
LoDResetKernel
<
float
,
XPUCtx
>
,
ops
::
LoDResetKernel
<
double
,
XPUCtx
>
,
ops
::
LoDResetKernel
<
int
,
XPUCtx
>
,
ops
::
LoDResetKernel
<
int64_t
,
XPUCtx
>
);
#endif
REGISTER_OP_CPU_KERNEL
(
lod_reset_grad
,
ops
::
LoDResetGradKernel
<
paddle
::
platform
::
CPUPlace
,
paddle
::
platform
::
float16
>
,
ops
::
LoDResetGradKernel
<
paddle
::
platform
::
CPUPlace
,
float
>
,
ops
::
LoDResetGradKernel
<
paddle
::
platform
::
CPUPlace
,
double
>
,
ops
::
LoDResetGradKernel
<
paddle
::
platform
::
CPUPlace
,
int
>
,
ops
::
LoDResetGradKernel
<
paddle
::
platform
::
CPUPlace
,
int64_t
>
);
PD_REGISTER_STRUCT_KERNEL
(
lod_reset_grad
,
CPU
,
ALL_LAYOUT
,
ops
::
LoDResetGradKernel
,
plat
::
float16
,
float
,
double
,
int
,
int64_t
)
{}
paddle/fluid/operators/lod_reset_op.cu
浏览文件 @
37ca3b4c
...
...
@@ -16,13 +16,19 @@ limitations under the License. */
namespace
ops
=
paddle
::
operators
;
REGISTER_OP_CUDA_KERNEL
(
lod_reset
,
ops
::
LoDResetKernel
<
phi
::
GPUContext
,
float
>
,
ops
::
LoDResetKernel
<
phi
::
GPUContext
,
double
>
,
ops
::
LoDResetKernel
<
phi
::
GPUContext
,
int
>
,
ops
::
LoDResetKernel
<
phi
::
GPUContext
,
int64_t
>
);
REGISTER_OP_CUDA_KERNEL
(
lod_reset_grad
,
ops
::
LoDResetGradKernel
<
phi
::
GPUContext
,
float
>
,
ops
::
LoDResetGradKernel
<
phi
::
GPUContext
,
double
>
,
ops
::
LoDResetGradKernel
<
phi
::
GPUContext
,
int
>
,
ops
::
LoDResetGradKernel
<
phi
::
GPUContext
,
int64_t
>
);
PD_REGISTER_STRUCT_KERNEL
(
lod_reset
,
GPU
,
ALL_LAYOUT
,
ops
::
LoDResetKernel
,
float
,
double
,
int
,
int64_t
)
{}
PD_REGISTER_STRUCT_KERNEL
(
lod_reset_grad
,
GPU
,
ALL_LAYOUT
,
ops
::
LoDResetGradKernel
,
float
,
double
,
int
,
int64_t
)
{}
paddle/fluid/operators/lod_reset_op.h
浏览文件 @
37ca3b4c
...
...
@@ -35,7 +35,7 @@ limitations under the License. */
namespace
paddle
{
namespace
operators
{
template
<
typename
DeviceContext
,
typename
T
>
template
<
typename
T
,
typename
DeviceContext
>
class
LoDResetKernel
:
public
framework
::
OpKernel
<
T
>
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
ctx
)
const
{
...
...
@@ -123,7 +123,7 @@ class LoDResetKernel : public framework::OpKernel<T> {
}
};
template
<
typename
DeviceContext
,
typename
T
>
template
<
typename
T
,
typename
DeviceContext
>
class
LoDResetGradKernel
:
public
framework
::
OpKernel
<
T
>
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
ctx
)
const
{
...
...
paddle/fluid/operators/lookup_table_dequant_op.cc
浏览文件 @
37ca3b4c
...
...
@@ -133,5 +133,9 @@ REGISTER_OPERATOR(
ops
::
LookupTableDequantOpMaker
,
paddle
::
framework
::
EmptyGradOpMaker
<
paddle
::
framework
::
OpDesc
>
,
paddle
::
framework
::
EmptyGradOpMaker
<
paddle
::
imperative
::
OpBase
>
);
REGISTER_OP_CPU_KERNEL
(
lookup_table_dequant
,
ops
::
LookupTableDequantKernel
<
float
>
);
PD_REGISTER_STRUCT_KERNEL
(
lookup_table_dequant
,
CPU
,
ALL_LAYOUT
,
ops
::
LookupTableDequantKernel
,
float
)
{}
paddle/fluid/operators/lookup_table_dequant_op.h
浏览文件 @
37ca3b4c
...
...
@@ -46,7 +46,7 @@ void dequant(const unsigned char *in,
constexpr
int64_t
kNoPadding
=
-
1
;
template
<
typename
T
>
template
<
typename
T
,
typename
DeviceContext
>
class
LookupTableDequantKernel
:
public
framework
::
OpKernel
<
T
>
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
context
)
const
override
{
...
...
paddle/fluid/operators/marker_op.cc
浏览文件 @
37ca3b4c
...
...
@@ -54,7 +54,7 @@ class MarkerOpMaker : public framework::OpProtoAndCheckerMaker {
}
};
template
<
typename
T
>
template
<
typename
T
,
typename
DeviceContext
>
class
MarkerOpCPUKernel
:
public
framework
::
OpKernel
<
T
>
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
ctx
)
const
override
{
...
...
@@ -75,4 +75,5 @@ class MarkerOpCPUKernel : public framework::OpKernel<T> {
namespace
ops
=
paddle
::
operators
;
REGISTER_OP_WITHOUT_GRADIENT
(
marker
,
ops
::
MarkerOp
,
ops
::
MarkerOpMaker
);
REGISTER_OP_CPU_KERNEL
(
marker
,
ops
::
MarkerOpCPUKernel
<
float
>
);
PD_REGISTER_STRUCT_KERNEL
(
marker
,
CPU
,
ALL_LAYOUT
,
ops
::
MarkerOpCPUKernel
,
float
)
{}
paddle/fluid/operators/marker_op.cu
浏览文件 @
37ca3b4c
...
...
@@ -29,7 +29,7 @@ __global__ void SimpleMarkerKernel(T* in, T* out, int ndim) {
}
}
template
<
typename
T
>
template
<
typename
T
,
typename
DeviceContext
>
class
MarkerOpCUDAKernel
:
public
framework
::
OpKernel
<
T
>
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
ctx
)
const
override
{
...
...
@@ -61,4 +61,5 @@ class MarkerOpCUDAKernel : public framework::OpKernel<T> {
namespace
ops
=
paddle
::
operators
;
namespace
plat
=
paddle
::
platform
;
REGISTER_OP_CUDA_KERNEL
(
marker
,
ops
::
MarkerOpCUDAKernel
<
float
>
);
PD_REGISTER_STRUCT_KERNEL
(
marker
,
GPU
,
ALL_LAYOUT
,
ops
::
MarkerOpCUDAKernel
,
float
)
{}
paddle/fluid/operators/save_load_combine_op_test.cc
浏览文件 @
37ca3b4c
...
...
@@ -23,8 +23,9 @@ limitations under the License. */
#include "paddle/phi/core/kernel_registry.h"
USE_OP_ITSELF
(
save_combine
);
USE_OP_ITSELF
(
load_combine
);
PD_DECLARE_KERNEL
(
save_combine_tensor
,
CPU
,
ALL_LAYOUT
);
USE_CPU_ONLY_OP
(
load_combine
);
PD_DECLARE_KERNEL
(
load_combine
,
CPU
,
ALL_LAYOUT
);
template
<
typename
T
,
typename
U
>
T
*
CreateForSaveCombineOp
(
int
x
,
...
...
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