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
兽拳
Paddle
提交
2f74e608
P
Paddle
项目概览
兽拳
/
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
2f74e608
编写于
8月 14, 2017
作者:
Q
qiaolongfei
浏览文件
操作
浏览文件
下载
差异文件
Merge branch 'develop' of
https://github.com/PaddlePaddle/Paddle
into public_to_protected
上级
d2c2f785
0c3c609d
变更
35
隐藏空白更改
内联
并排
Showing
35 changed file
with
614 addition
and
112 deletion
+614
-112
paddle/framework/backward_test.cc
paddle/framework/backward_test.cc
+4
-3
paddle/framework/grad_op_builder.cc
paddle/framework/grad_op_builder.cc
+20
-20
paddle/framework/grad_op_builder_test.cc
paddle/framework/grad_op_builder_test.cc
+1
-1
paddle/framework/op_registry.h
paddle/framework/op_registry.h
+3
-5
paddle/framework/op_registry_test.cc
paddle/framework/op_registry_test.cc
+2
-2
paddle/framework/operator.cc
paddle/framework/operator.cc
+17
-0
paddle/framework/operator.h
paddle/framework/operator.h
+9
-18
paddle/framework/operator_test.cc
paddle/framework/operator_test.cc
+7
-5
paddle/framework/tensor.h
paddle/framework/tensor.h
+2
-0
paddle/operators/CMakeLists.txt
paddle/operators/CMakeLists.txt
+2
-1
paddle/operators/add_op.cc
paddle/operators/add_op.cc
+5
-2
paddle/operators/cross_entropy_op.cc
paddle/operators/cross_entropy_op.cc
+6
-3
paddle/operators/fill_zeros_like_op.cc
paddle/operators/fill_zeros_like_op.cc
+2
-1
paddle/operators/gaussian_random_op.cc
paddle/operators/gaussian_random_op.cc
+2
-1
paddle/operators/math/CMakeLists.txt
paddle/operators/math/CMakeLists.txt
+13
-0
paddle/operators/math/math_function.cc
paddle/operators/math/math_function.cc
+114
-0
paddle/operators/math/math_function.cu
paddle/operators/math/math_function.cu
+127
-0
paddle/operators/math/math_function.h
paddle/operators/math/math_function.h
+82
-0
paddle/operators/math/math_function_test.cc
paddle/operators/math/math_function_test.cc
+75
-0
paddle/operators/mean_op.cc
paddle/operators/mean_op.cc
+6
-2
paddle/operators/mul_op.cc
paddle/operators/mul_op.cc
+6
-2
paddle/operators/mul_op.cu
paddle/operators/mul_op.cu
+0
-1
paddle/operators/mul_op.h
paddle/operators/mul_op.h
+3
-0
paddle/operators/net_op.cc
paddle/operators/net_op.cc
+6
-0
paddle/operators/net_op.h
paddle/operators/net_op.h
+3
-1
paddle/operators/net_op_test.cc
paddle/operators/net_op_test.cc
+11
-11
paddle/operators/recurrent_op.cc
paddle/operators/recurrent_op.cc
+10
-4
paddle/operators/recurrent_op.h
paddle/operators/recurrent_op.h
+7
-8
paddle/operators/rowwise_add_op.cc
paddle/operators/rowwise_add_op.cc
+3
-1
paddle/operators/sgd_op.cc
paddle/operators/sgd_op.cc
+3
-1
paddle/operators/sigmoid_op.cc
paddle/operators/sigmoid_op.cc
+6
-2
paddle/operators/softmax_op.cc
paddle/operators/softmax_op.cc
+6
-2
paddle/operators/uniform_random_op.cc
paddle/operators/uniform_random_op.cc
+3
-1
paddle/platform/dynload/cublas.h
paddle/platform/dynload/cublas.h
+6
-6
paddle/platform/enforce.h
paddle/platform/enforce.h
+42
-8
未找到文件。
paddle/framework/backward_test.cc
浏览文件 @
2f74e608
...
@@ -30,7 +30,7 @@ using DeviceContext = platform::DeviceContext;
...
@@ -30,7 +30,7 @@ using DeviceContext = platform::DeviceContext;
class
EmptyOp
:
public
OperatorBase
{
class
EmptyOp
:
public
OperatorBase
{
public:
public:
DEFINE_OPERATOR_CTOR
(
EmptyOp
,
OperatorBase
)
;
using
OperatorBase
::
OperatorBase
;
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
void
Run
(
const
Scope
&
scope
,
const
DeviceContext
&
dev_ctx
)
const
override
{}
void
Run
(
const
Scope
&
scope
,
const
DeviceContext
&
dev_ctx
)
const
override
{}
};
};
...
@@ -79,8 +79,9 @@ class NoGradOpMaker : public OpProtoAndCheckerMaker {
...
@@ -79,8 +79,9 @@ class NoGradOpMaker : public OpProtoAndCheckerMaker {
class
FcOp
:
public
operators
::
NetOp
{
class
FcOp
:
public
operators
::
NetOp
{
public:
public:
DEFINE_OPERATOR_CTOR
(
FcOp
,
operators
::
NetOp
)
FcOp
(
const
std
::
string
&
type
,
const
VarNameMap
&
inputs
,
void
Init
()
override
{
const
VarNameMap
&
outputs
,
const
AttributeMap
&
attrs
)
:
NetOp
(
type
,
inputs
,
outputs
,
attrs
)
{
AddOp
(
OpRegistry
::
CreateOp
(
"mul"
,
AddOp
(
OpRegistry
::
CreateOp
(
"mul"
,
{{
"X"
,
{
Input
(
"X"
)}},
{
"Y"
,
{
Input
(
"W"
)}}},
{{
"X"
,
{
Input
(
"X"
)}},
{
"Y"
,
{
Input
(
"W"
)}}},
{{
"Out"
,
{
Output
(
"mul_result"
)}}},
{}));
{{
"Out"
,
{
Output
(
"mul_result"
)}}},
{}));
...
...
paddle/framework/grad_op_builder.cc
浏览文件 @
2f74e608
...
@@ -19,14 +19,12 @@ namespace paddle {
...
@@ -19,14 +19,12 @@ namespace paddle {
namespace
framework
{
namespace
framework
{
enum
class
OpArgType
{
IN
,
OUT
};
enum
class
OpArgType
{
IN
,
OUT
};
using
VarNameMap
=
OperatorBase
::
VarNameMap
;
static
void
TransOpArg
(
const
OperatorBase
*
src_op
,
OperatorBase
::
VarNameMap
*
vars
,
static
VarNameMap
TransOpArg
(
const
OperatorBase
*
src_op
,
const
OpArgType
&
src_type
,
bool
is_grad
)
{
const
OpArgType
&
src_type
,
const
OpArgType
&
dst_type
,
bool
is_grad
)
{
const
auto
&
src_inout
=
const
auto
&
src_inout
=
src_type
==
OpArgType
::
IN
?
src_op
->
Inputs
()
:
src_op
->
Outputs
()
;
src_type
==
OpArgType
::
IN
?
src_op
->
inputs_
:
src_op
->
outputs_
;
VarNameMap
dst_inout
;
auto
&
dst_inout
=
*
vars
;
const
OpProto
&
proto
=
OpProtos
().
at
(
src_op
->
Type
());
const
OpProto
&
proto
=
OpProtos
().
at
(
src_op
->
Type
());
const
auto
&
src_arg_list
=
const
auto
&
src_arg_list
=
...
@@ -45,20 +43,22 @@ static VarNameMap TransOpArg(const OperatorBase* src_op,
...
@@ -45,20 +43,22 @@ static VarNameMap TransOpArg(const OperatorBase* src_op,
}
}
OperatorBase
*
BuildGradOp
(
const
OperatorBase
*
op
)
{
OperatorBase
*
BuildGradOp
(
const
OperatorBase
*
op
)
{
std
::
string
grad_op_type
=
OpRegistry
::
grad_ops
().
at
(
op
->
Type
());
auto
gop_type_it
=
OpRegistry
::
grad_ops
().
find
(
op
->
type_
);
auto
I
=
TransOpArg
(
op
,
OpArgType
::
IN
,
OpArgType
::
IN
,
false
);
// I
PADDLE_ENFORCE
(
gop_type_it
!=
OpRegistry
::
grad_ops
().
end
(),
auto
O
=
TransOpArg
(
op
,
OpArgType
::
OUT
,
OpArgType
::
IN
,
false
);
// O
"Operator %s do not register gradient type"
,
op
->
type_
);
auto
OG
=
TransOpArg
(
op
,
OpArgType
::
OUT
,
OpArgType
::
IN
,
true
);
// OG
auto
&
grad_op_type
=
gop_type_it
->
second
;
auto
IG
=
TransOpArg
(
op
,
OpArgType
::
IN
,
OpArgType
::
OUT
,
true
);
// IG
OperatorBase
::
VarNameMap
inputs
;
// TODO(merge I/O/OG)
OperatorBase
::
VarNameMap
outputs
;
VarNameMap
GradIn
;
TransOpArg
(
op
,
&
inputs
,
OpArgType
::
IN
,
false
);
// I
GradIn
.
insert
(
I
.
begin
(),
I
.
end
());
TransOpArg
(
op
,
&
inputs
,
OpArgType
::
OUT
,
false
);
// O
GradIn
.
insert
(
O
.
begin
(),
O
.
end
());
TransOpArg
(
op
,
&
inputs
,
OpArgType
::
OUT
,
true
);
// OG
GradIn
.
insert
(
OG
.
begin
(),
OG
.
end
());
TransOpArg
(
op
,
&
outputs
,
OpArgType
::
IN
,
true
);
// IG
auto
gop_it
=
OpRegistry
::
op_creators
().
find
(
grad_op_type
);
PADDLE_ENFORCE
(
gop_it
!=
OpRegistry
::
op_creators
().
end
(),
"Operator %s 's Gradient %s's creator cannot be found"
,
op
->
type_
,
grad_op_type
);
OperatorBase
*
grad_op
=
OpRegistry
::
op_creators
().
at
(
grad_op_type
)(
return
gop_it
->
second
(
grad_op_type
,
inputs
,
outputs
,
op
->
attrs_
);
grad_op_type
,
GradIn
,
IG
,
op
->
Attrs
());
return
grad_op
;
}
}
}
// namespace framework
}
// namespace framework
...
...
paddle/framework/grad_op_builder_test.cc
浏览文件 @
2f74e608
...
@@ -10,7 +10,7 @@ namespace framework {
...
@@ -10,7 +10,7 @@ namespace framework {
class
NOP
:
public
OperatorBase
{
class
NOP
:
public
OperatorBase
{
public:
public:
DEFINE_OPERATOR_CTOR
(
NOP
,
OperatorBase
)
;
using
OperatorBase
::
OperatorBase
;
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
void
Run
(
const
Scope
&
scope
,
void
Run
(
const
Scope
&
scope
,
const
platform
::
DeviceContext
&
dev_ctx
)
const
override
{}
const
platform
::
DeviceContext
&
dev_ctx
)
const
override
{}
...
...
paddle/framework/op_registry.h
浏览文件 @
2f74e608
...
@@ -122,8 +122,8 @@ class OpProtoAndCheckerMaker {
...
@@ -122,8 +122,8 @@ class OpProtoAndCheckerMaker {
class
OpRegistry
{
class
OpRegistry
{
using
VarNameMap
=
OperatorBase
::
VarNameMap
;
using
VarNameMap
=
OperatorBase
::
VarNameMap
;
using
OpCreator
=
std
::
function
<
OperatorBase
*
(
using
OpCreator
=
std
::
function
<
OperatorBase
*
(
const
std
::
string
&
type
,
const
VarNameMap
&
inputs
,
const
std
::
string
&
/*type*/
,
const
VarNameMap
&
/*inputs*/
,
const
VarNameMap
&
outputs
,
const
AttributeMap
&
attrs
)
>
;
const
VarNameMap
&
/*outputs*/
,
const
AttributeMap
&
/*attrs*/
)
>
;
public:
public:
template
<
typename
OpType
,
typename
ProtoMakerType
>
template
<
typename
OpType
,
typename
ProtoMakerType
>
...
@@ -158,7 +158,7 @@ class OpRegistry {
...
@@ -158,7 +158,7 @@ class OpRegistry {
static
std
::
shared_ptr
<
OperatorBase
>
CreateOp
(
const
std
::
string
&
type
,
static
std
::
shared_ptr
<
OperatorBase
>
CreateOp
(
const
std
::
string
&
type
,
const
VarNameMap
&
inputs
,
const
VarNameMap
&
inputs
,
const
VarNameMap
&
outputs
,
const
VarNameMap
&
outputs
,
const
AttributeMap
&
attrs
)
{
AttributeMap
attrs
)
{
auto
op_create_it
=
op_creators
().
find
(
type
);
auto
op_create_it
=
op_creators
().
find
(
type
);
PADDLE_ENFORCE
(
op_create_it
!=
op_creators
().
end
(),
PADDLE_ENFORCE
(
op_create_it
!=
op_creators
().
end
(),
"Operator %s cannot be found."
,
type
);
"Operator %s cannot be found."
,
type
);
...
@@ -168,7 +168,6 @@ class OpRegistry {
...
@@ -168,7 +168,6 @@ class OpRegistry {
auto
op
=
op_create_it
->
second
(
type
,
inputs
,
outputs
,
attrMap
);
auto
op
=
op_create_it
->
second
(
type
,
inputs
,
outputs
,
attrMap
);
GenerateTempVariableName
(
op
);
GenerateTempVariableName
(
op
);
op
->
Init
();
return
std
::
shared_ptr
<
OperatorBase
>
(
op
);
return
std
::
shared_ptr
<
OperatorBase
>
(
op
);
}
}
...
@@ -200,7 +199,6 @@ class OpRegistry {
...
@@ -200,7 +199,6 @@ class OpRegistry {
PADDLE_ENFORCE
(
!
op
.
IsNetOp
(),
PADDLE_ENFORCE
(
!
op
.
IsNetOp
(),
"Use framework::Backward to get backward ops"
);
"Use framework::Backward to get backward ops"
);
std
::
shared_ptr
<
OperatorBase
>
grad_op
(
BuildGradOp
(
&
op
));
std
::
shared_ptr
<
OperatorBase
>
grad_op
(
BuildGradOp
(
&
op
));
grad_op
->
Init
();
return
grad_op
;
return
grad_op
;
}
}
...
...
paddle/framework/op_registry_test.cc
浏览文件 @
2f74e608
...
@@ -7,7 +7,7 @@ namespace paddle {
...
@@ -7,7 +7,7 @@ namespace paddle {
namespace
framework
{
namespace
framework
{
class
CosineOp
:
public
OperatorBase
{
class
CosineOp
:
public
OperatorBase
{
public:
public:
DEFINE_OPERATOR_CTOR
(
CosineOp
,
OperatorBase
)
;
using
OperatorBase
::
OperatorBase
;
void
Run
(
const
Scope
&
scope
,
void
Run
(
const
Scope
&
scope
,
const
platform
::
DeviceContext
&
dev_ctx
)
const
override
{}
const
platform
::
DeviceContext
&
dev_ctx
)
const
override
{}
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
...
@@ -28,7 +28,7 @@ class CosineOpProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
...
@@ -28,7 +28,7 @@ class CosineOpProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
class
MyTestOp
:
public
OperatorBase
{
class
MyTestOp
:
public
OperatorBase
{
public:
public:
DEFINE_OPERATOR_CTOR
(
MyTestOp
,
OperatorBase
)
;
using
OperatorBase
::
OperatorBase
;
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
void
Run
(
const
Scope
&
scope
,
void
Run
(
const
Scope
&
scope
,
const
platform
::
DeviceContext
&
dev_ctx
)
const
override
{}
const
platform
::
DeviceContext
&
dev_ctx
)
const
override
{}
...
...
paddle/framework/operator.cc
浏览文件 @
2f74e608
...
@@ -122,6 +122,23 @@ void OperatorBase::Rename(const std::string& old_name,
...
@@ -122,6 +122,23 @@ void OperatorBase::Rename(const std::string& old_name,
}
}
}
}
OperatorBase
::
OperatorBase
(
const
std
::
string
&
type
,
const
OperatorBase
::
VarNameMap
&
inputs
,
const
OperatorBase
::
VarNameMap
&
outputs
,
const
AttributeMap
&
attrs
)
:
type_
(
type
),
inputs_
(
inputs
),
outputs_
(
outputs
),
attrs_
(
attrs
)
{
static
std
::
atomic
<
size_t
>
gUniqId
(
0UL
);
for
(
auto
&
output
:
outputs_
)
{
for
(
auto
&
output_name
:
output
.
second
)
{
if
(
output_name
==
kTempVarName
)
{
output_name
+=
type_
;
output_name
+=
"@"
;
output_name
+=
std
::
to_string
(
gUniqId
.
fetch_add
(
1
));
}
}
}
}
std
::
vector
<
std
::
string
>
OperatorBase
::
OutputVars
(
bool
has_intermediate
)
const
{
std
::
vector
<
std
::
string
>
OperatorBase
::
OutputVars
(
bool
has_intermediate
)
const
{
std
::
vector
<
std
::
string
>
ret_val
;
std
::
vector
<
std
::
string
>
ret_val
;
if
(
has_intermediate
)
{
if
(
has_intermediate
)
{
...
...
paddle/framework/operator.h
浏览文件 @
2f74e608
...
@@ -66,10 +66,8 @@ class OperatorBase {
...
@@ -66,10 +66,8 @@ class OperatorBase {
public:
public:
using
VarNameMap
=
std
::
map
<
std
::
string
,
std
::
vector
<
std
::
string
>>
;
using
VarNameMap
=
std
::
map
<
std
::
string
,
std
::
vector
<
std
::
string
>>
;
OperatorBase
()
=
default
;
OperatorBase
(
const
std
::
string
&
type
,
const
VarNameMap
&
inputs
,
OperatorBase
(
const
std
::
string
&
type
,
const
VarNameMap
&
inputs
,
const
VarNameMap
&
outputs
,
const
AttributeMap
&
attrs
)
const
VarNameMap
&
outputs
,
const
AttributeMap
&
attrs
);
:
type_
(
type
),
inputs_
(
inputs
),
outputs_
(
outputs
),
attrs_
(
attrs
)
{}
OperatorBase
(
const
OperatorBase
&
o
)
=
delete
;
OperatorBase
(
const
OperatorBase
&
o
)
=
delete
;
OperatorBase
&
operator
=
(
const
OperatorBase
&
o
)
=
delete
;
OperatorBase
&
operator
=
(
const
OperatorBase
&
o
)
=
delete
;
...
@@ -86,10 +84,6 @@ class OperatorBase {
...
@@ -86,10 +84,6 @@ class OperatorBase {
virtual
std
::
string
DebugString
()
const
;
virtual
std
::
string
DebugString
()
const
;
/// Init will be called after CreateOperator, you can put some initialization
/// logic here.
virtual
void
Init
()
{}
/// InferShape infer the size of Variables used by this Operator with
/// InferShape infer the size of Variables used by this Operator with
/// information inside scope
/// information inside scope
virtual
void
InferShape
(
const
Scope
&
scope
)
const
=
0
;
virtual
void
InferShape
(
const
Scope
&
scope
)
const
=
0
;
...
@@ -138,15 +132,6 @@ class OperatorBase {
...
@@ -138,15 +132,6 @@ class OperatorBase {
AttributeMap
attrs_
;
AttributeMap
attrs_
;
};
};
#define DEFINE_OPERATOR_CTOR(Class, ParentClass) \
public: \
Class() : ParentClass() {
/* TODO(yi): This constructor is to be removed. */
\
} \
Class(const std::string& type, const VarNameMap& inputs, \
const VarNameMap& outputs, \
const paddle::framework::AttributeMap& attrs) \
: ParentClass(type, inputs, outputs, attrs) {}
class
InferShapeContext
{
class
InferShapeContext
{
public:
public:
InferShapeContext
(
const
OperatorBase
&
op
,
const
Scope
&
scope
)
InferShapeContext
(
const
OperatorBase
&
op
,
const
Scope
&
scope
)
...
@@ -267,6 +252,10 @@ class ExecutionContext : public InferShapeContext {
...
@@ -267,6 +252,10 @@ class ExecutionContext : public InferShapeContext {
platform
::
Place
GetPlace
()
const
{
return
device_context_
->
GetPlace
();
}
platform
::
Place
GetPlace
()
const
{
return
device_context_
->
GetPlace
();
}
const
platform
::
DeviceContext
*
device_context
()
const
{
return
device_context_
;
}
const
platform
::
DeviceContext
*
device_context_
;
const
platform
::
DeviceContext
*
device_context_
;
};
};
...
@@ -286,8 +275,6 @@ class OpKernel {
...
@@ -286,8 +275,6 @@ class OpKernel {
class
OperatorWithKernel
:
public
OperatorBase
{
class
OperatorWithKernel
:
public
OperatorBase
{
public:
public:
DEFINE_OPERATOR_CTOR
(
OperatorWithKernel
,
OperatorBase
)
struct
OpKernelKey
{
struct
OpKernelKey
{
platform
::
Place
place_
;
platform
::
Place
place_
;
...
@@ -311,6 +298,10 @@ class OperatorWithKernel : public OperatorBase {
...
@@ -311,6 +298,10 @@ class OperatorWithKernel : public OperatorBase {
using
OpKernelMap
=
using
OpKernelMap
=
std
::
unordered_map
<
OpKernelKey
,
std
::
unique_ptr
<
OpKernel
>
,
OpKernelHash
>
;
std
::
unordered_map
<
OpKernelKey
,
std
::
unique_ptr
<
OpKernel
>
,
OpKernelHash
>
;
OperatorWithKernel
(
const
std
::
string
&
type
,
const
VarNameMap
&
inputs
,
const
VarNameMap
&
outputs
,
const
AttributeMap
&
attrs
)
:
OperatorBase
(
type
,
inputs
,
outputs
,
attrs
)
{}
void
InferShape
(
const
Scope
&
scope
)
const
override
{
void
InferShape
(
const
Scope
&
scope
)
const
override
{
InferShape
(
InferShapeContext
(
*
this
,
scope
));
InferShape
(
InferShapeContext
(
*
this
,
scope
));
}
}
...
...
paddle/framework/operator_test.cc
浏览文件 @
2f74e608
...
@@ -22,10 +22,10 @@ namespace framework {
...
@@ -22,10 +22,10 @@ namespace framework {
static
int
op_run_num
=
0
;
static
int
op_run_num
=
0
;
class
OpWithoutKernelTest
:
public
OperatorBase
{
class
OpWithoutKernelTest
:
public
OperatorBase
{
DEFINE_OPERATOR_CTOR
(
OpWithoutKernelTest
,
framework
::
OperatorBase
)
public:
public:
void
Init
()
override
{
x
=
1
;
}
OpWithoutKernelTest
(
const
std
::
string
&
type
,
const
VarNameMap
&
inputs
,
const
VarNameMap
&
outputs
,
const
AttributeMap
&
attrs
)
:
OperatorBase
(
type
,
inputs
,
outputs
,
attrs
),
x
(
1
)
{}
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
void
Run
(
const
Scope
&
scope
,
void
Run
(
const
Scope
&
scope
,
const
platform
::
DeviceContext
&
dev_ctx
)
const
override
{
const
platform
::
DeviceContext
&
dev_ctx
)
const
override
{
...
@@ -38,7 +38,7 @@ class OpWithoutKernelTest : public OperatorBase {
...
@@ -38,7 +38,7 @@ class OpWithoutKernelTest : public OperatorBase {
}
}
public:
public:
float
x
=
0
;
int
x
{
0
}
;
};
};
class
OpeWithoutKernelTestProtoAndCheckerMaker
:
public
OpProtoAndCheckerMaker
{
class
OpeWithoutKernelTestProtoAndCheckerMaker
:
public
OpProtoAndCheckerMaker
{
...
@@ -109,7 +109,9 @@ class OpKernelTestProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
...
@@ -109,7 +109,9 @@ class OpKernelTestProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
static
int
cpu_kernel_run_num
=
0
;
static
int
cpu_kernel_run_num
=
0
;
class
OpWithKernelTest
:
public
OperatorWithKernel
{
class
OpWithKernelTest
:
public
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
OpWithKernelTest
,
framework
::
OperatorWithKernel
)
public:
using
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{}
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{}
};
};
...
...
paddle/framework/tensor.h
浏览文件 @
2f74e608
...
@@ -105,6 +105,8 @@ class Tensor {
...
@@ -105,6 +105,8 @@ class Tensor {
template
<
typename
T
>
template
<
typename
T
>
inline
Tensor
Slice
(
const
int
&
begin_idx
,
const
int
&
end_idx
)
const
;
inline
Tensor
Slice
(
const
int
&
begin_idx
,
const
int
&
end_idx
)
const
;
platform
::
Place
place
()
const
{
return
holder_
->
place
();
}
private:
private:
template
<
typename
T
>
template
<
typename
T
>
inline
void
check_memory_size
()
const
;
inline
void
check_memory_size
()
const
;
...
...
paddle/operators/CMakeLists.txt
浏览文件 @
2f74e608
...
@@ -41,6 +41,7 @@ function(op_library TARGET)
...
@@ -41,6 +41,7 @@ function(op_library TARGET)
endif
()
endif
()
endfunction
()
endfunction
()
add_subdirectory
(
math
)
cc_test
(
gather_test SRCS gather_test.cc DEPS tensor
)
cc_test
(
gather_test SRCS gather_test.cc DEPS tensor
)
cc_library
(
net_op SRCS net_op.cc DEPS op_registry
)
cc_library
(
net_op SRCS net_op.cc DEPS op_registry
)
...
@@ -50,7 +51,7 @@ op_library(add_op SRCS add_op.cc add_op.cu)
...
@@ -50,7 +51,7 @@ op_library(add_op SRCS add_op.cc add_op.cu)
op_library
(
mean_op SRCS mean_op.cc mean_op.cu
)
op_library
(
mean_op SRCS mean_op.cc mean_op.cu
)
op_library
(
mul_op SRCS mul_op.cc mul_op.cu
)
op_library
(
mul_op SRCS mul_op.cc mul_op.cu
DEPS math_function
)
op_library
(
rowwise_add_op SRCS rowwise_add_op.cu rowwise_add_op.cc
)
op_library
(
rowwise_add_op SRCS rowwise_add_op.cu rowwise_add_op.cc
)
op_library
(
sigmoid_op SRCS sigmoid_op.cc sigmoid_op.cu
)
op_library
(
sigmoid_op SRCS sigmoid_op.cc sigmoid_op.cu
)
...
...
paddle/operators/add_op.cc
浏览文件 @
2f74e608
...
@@ -18,7 +18,8 @@ namespace paddle {
...
@@ -18,7 +18,8 @@ namespace paddle {
namespace
operators
{
namespace
operators
{
class
AddOp
:
public
framework
::
OperatorWithKernel
{
class
AddOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
AddOp
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
...
@@ -45,7 +46,9 @@ The equation is: Out = X + Y
...
@@ -45,7 +46,9 @@ The equation is: Out = X + Y
};
};
class
AddOpGrad
:
public
framework
::
OperatorWithKernel
{
class
AddOpGrad
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
AddOpGrad
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{}
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{}
};
};
...
...
paddle/operators/cross_entropy_op.cc
浏览文件 @
2f74e608
...
@@ -18,7 +18,9 @@ namespace paddle {
...
@@ -18,7 +18,9 @@ namespace paddle {
namespace
operators
{
namespace
operators
{
class
OnehotCrossEntropyOp
:
public
framework
::
OperatorWithKernel
{
class
OnehotCrossEntropyOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
OnehotCrossEntropyOp
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
auto
*
X
=
ctx
.
Input
<
Tensor
>
(
"X"
);
auto
*
X
=
ctx
.
Input
<
Tensor
>
(
"X"
);
...
@@ -32,8 +34,9 @@ class OnehotCrossEntropyOp : public framework::OperatorWithKernel {
...
@@ -32,8 +34,9 @@ class OnehotCrossEntropyOp : public framework::OperatorWithKernel {
};
};
class
OnehotCrossEntropyGradientOp
:
public
framework
::
OperatorWithKernel
{
class
OnehotCrossEntropyGradientOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
OnehotCrossEntropyGradientOp
,
public:
framework
::
OperatorWithKernel
)
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
auto
X_grad
=
ctx
.
Output
<
Tensor
>
(
framework
::
GradVarName
(
"X"
));
auto
X_grad
=
ctx
.
Output
<
Tensor
>
(
framework
::
GradVarName
(
"X"
));
...
...
paddle/operators/fill_zeros_like_op.cc
浏览文件 @
2f74e608
...
@@ -18,7 +18,8 @@ namespace paddle {
...
@@ -18,7 +18,8 @@ namespace paddle {
namespace
operators
{
namespace
operators
{
class
FillZerosLikeOp
:
public
framework
::
OperatorWithKernel
{
class
FillZerosLikeOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
FillZerosLikeOp
,
framework
::
OperatorWithKernel
);
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
...
...
paddle/operators/gaussian_random_op.cc
浏览文件 @
2f74e608
...
@@ -43,7 +43,8 @@ class GaussianRandomKernel : public framework::OpKernel {
...
@@ -43,7 +43,8 @@ class GaussianRandomKernel : public framework::OpKernel {
};
};
class
GaussianRandomOp
:
public
framework
::
OperatorWithKernel
{
class
GaussianRandomOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
GaussianRandomOp
,
framework
::
OperatorWithKernel
);
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
context
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
context
)
const
override
{
...
...
paddle/operators/math/CMakeLists.txt
0 → 100644
浏览文件 @
2f74e608
if
(
WITH_MKLML
)
set
(
BLAS_LIB mklml
)
else
()
set
(
BLAS_LIB cblas
)
endif
()
if
(
WITH_GPU
)
nv_library
(
math_function SRCS math_function.cc math_function.cu DEPS
${
BLAS_LIB
}
device_context
)
else
()
cc_library
(
math_function SRCS math_function.cc DEPS
${
BLAS_LIB
}
device_context
)
endif
()
nv_test
(
math_function_test SRCS math_function_test.cc DEPS math_function tensor
)
paddle/operators/math/math_function.cc
0 → 100644
浏览文件 @
2f74e608
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
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/operators/math/math_function.h"
namespace
paddle
{
namespace
operators
{
namespace
math
{
template
<
>
void
gemm
<
platform
::
CPUPlace
,
float
>
(
const
CBLAS_TRANSPOSE
transA
,
const
CBLAS_TRANSPOSE
transB
,
const
int
M
,
const
int
N
,
const
int
K
,
const
float
alpha
,
const
float
*
A
,
const
float
*
B
,
const
float
beta
,
float
*
C
,
platform
::
DeviceContext
*
context
)
{
int
lda
=
K
;
int
ldb
=
N
;
int
ldc
=
N
;
cblas_sgemm
(
CblasRowMajor
,
transA
,
transB
,
M
,
N
,
K
,
alpha
,
A
,
lda
,
B
,
ldb
,
beta
,
C
,
ldc
);
}
template
<
>
void
gemm
<
platform
::
CPUPlace
,
double
>
(
const
CBLAS_TRANSPOSE
transA
,
const
CBLAS_TRANSPOSE
transB
,
const
int
M
,
const
int
N
,
const
int
K
,
const
double
alpha
,
const
double
*
A
,
const
double
*
B
,
const
double
beta
,
double
*
C
,
platform
::
DeviceContext
*
context
)
{
int
lda
=
K
;
int
ldb
=
N
;
int
ldc
=
N
;
cblas_dgemm
(
CblasRowMajor
,
transA
,
transB
,
M
,
N
,
K
,
alpha
,
A
,
lda
,
B
,
ldb
,
beta
,
C
,
ldc
);
}
template
<
>
void
matmul
<
platform
::
CPUPlace
,
float
>
(
const
framework
::
Tensor
&
matrix_a
,
bool
trans_a
,
const
framework
::
Tensor
&
matrix_b
,
bool
trans_b
,
float
alpha
,
framework
::
Tensor
*
matrix_out
,
float
beta
,
platform
::
DeviceContext
*
context
)
{
auto
dim_a
=
matrix_a
.
dims
();
auto
dim_b
=
matrix_b
.
dims
();
auto
dim_out
=
matrix_out
->
dims
();
PADDLE_ENFORCE
(
dim_a
.
size
()
==
2
&&
dim_b
.
size
()
==
2
&&
dim_out
.
size
()
==
2
,
"The input and output of matmul be matrix"
);
PADDLE_ENFORCE
(
platform
::
is_cpu_place
(
matrix_a
.
place
())
&&
platform
::
is_cpu_place
(
matrix_b
.
place
())
&&
platform
::
is_cpu_place
(
matrix_out
->
place
()),
"Matrix must all be in CPUPlace"
);
int
M
=
dim_out
[
0
];
int
N
=
dim_out
[
1
];
int
K
=
(
trans_a
==
false
)
?
dim_a
[
1
]
:
dim_a
[
0
];
CBLAS_TRANSPOSE
transA
=
(
trans_a
==
false
)
?
CblasNoTrans
:
CblasTrans
;
CBLAS_TRANSPOSE
transB
=
(
trans_b
==
false
)
?
CblasNoTrans
:
CblasTrans
;
gemm
<
platform
::
CPUPlace
,
float
>
(
transA
,
transB
,
M
,
N
,
K
,
alpha
,
matrix_a
.
data
<
float
>
(),
matrix_b
.
data
<
float
>
(),
beta
,
matrix_out
->
data
<
float
>
(),
context
);
}
template
<
>
void
matmul
<
platform
::
CPUPlace
,
double
>
(
const
framework
::
Tensor
&
matrix_a
,
bool
trans_a
,
const
framework
::
Tensor
&
matrix_b
,
bool
trans_b
,
double
alpha
,
framework
::
Tensor
*
matrix_out
,
double
beta
,
platform
::
DeviceContext
*
context
)
{
auto
dim_a
=
matrix_a
.
dims
();
auto
dim_b
=
matrix_b
.
dims
();
auto
dim_out
=
matrix_out
->
dims
();
PADDLE_ENFORCE
(
dim_a
.
size
()
==
2
&&
dim_b
.
size
()
==
2
&&
dim_out
.
size
()
==
2
,
"The input and output of matmul be matrix"
);
PADDLE_ENFORCE
(
platform
::
is_cpu_place
(
matrix_a
.
place
())
&&
platform
::
is_cpu_place
(
matrix_b
.
place
())
&&
platform
::
is_cpu_place
(
matrix_out
->
place
()),
"Matrix must all be in CPUPlace"
);
int
M
=
dim_out
[
0
];
int
N
=
dim_out
[
1
];
int
K
=
(
trans_a
==
false
)
?
dim_a
[
1
]
:
dim_a
[
0
];
CBLAS_TRANSPOSE
transA
=
(
trans_a
==
false
)
?
CblasNoTrans
:
CblasTrans
;
CBLAS_TRANSPOSE
transB
=
(
trans_b
==
false
)
?
CblasNoTrans
:
CblasTrans
;
gemm
<
platform
::
CPUPlace
,
double
>
(
transA
,
transB
,
M
,
N
,
K
,
alpha
,
matrix_a
.
data
<
double
>
(),
matrix_b
.
data
<
double
>
(),
beta
,
matrix_out
->
data
<
double
>
(),
context
);
}
}
// namespace math
}
// namespace operators
}
// namespace paddle
paddle/operators/math/math_function.cu
0 → 100644
浏览文件 @
2f74e608
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
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/operators/math/math_function.h"
namespace
paddle
{
namespace
operators
{
namespace
math
{
template
<
>
void
gemm
<
platform
::
GPUPlace
,
float
>
(
const
CBLAS_TRANSPOSE
transA
,
const
CBLAS_TRANSPOSE
transB
,
const
int
M
,
const
int
N
,
const
int
K
,
const
float
alpha
,
const
float
*
A
,
const
float
*
B
,
const
float
beta
,
float
*
C
,
platform
::
DeviceContext
*
context
)
{
// Note that cublas follows fortran order, so the order is different from
// the cblas convention.
int
lda
=
(
transA
==
CblasNoTrans
)
?
K
:
M
;
int
ldb
=
(
transB
==
CblasNoTrans
)
?
N
:
K
;
cublasOperation_t
cuTransA
=
(
transA
==
CblasNoTrans
)
?
CUBLAS_OP_N
:
CUBLAS_OP_T
;
cublasOperation_t
cuTransB
=
(
transB
==
CblasNoTrans
)
?
CUBLAS_OP_N
:
CUBLAS_OP_T
;
PADDLE_ENFORCE
(
platform
::
dynload
::
cublasSgemm
(
reinterpret_cast
<
platform
::
CUDADeviceContext
*>
(
context
)
->
cublas_handle
(),
cuTransB
,
cuTransA
,
N
,
M
,
K
,
&
alpha
,
B
,
ldb
,
A
,
lda
,
&
beta
,
C
,
N
));
}
template
<
>
void
gemm
<
platform
::
GPUPlace
,
double
>
(
const
CBLAS_TRANSPOSE
transA
,
const
CBLAS_TRANSPOSE
transB
,
const
int
M
,
const
int
N
,
const
int
K
,
const
double
alpha
,
const
double
*
A
,
const
double
*
B
,
const
double
beta
,
double
*
C
,
platform
::
DeviceContext
*
context
)
{
// Note that cublas follows fortran order, so the order is different from
// the cblas convention.
int
lda
=
(
transA
==
CblasNoTrans
)
?
K
:
M
;
int
ldb
=
(
transB
==
CblasNoTrans
)
?
N
:
K
;
cublasOperation_t
cuTransA
=
(
transA
==
CblasNoTrans
)
?
CUBLAS_OP_N
:
CUBLAS_OP_T
;
cublasOperation_t
cuTransB
=
(
transB
==
CblasNoTrans
)
?
CUBLAS_OP_N
:
CUBLAS_OP_T
;
PADDLE_ENFORCE
(
platform
::
dynload
::
cublasDgemm
(
reinterpret_cast
<
platform
::
CUDADeviceContext
*>
(
context
)
->
cublas_handle
(),
cuTransB
,
cuTransA
,
N
,
M
,
K
,
&
alpha
,
B
,
ldb
,
A
,
lda
,
&
beta
,
C
,
N
));
}
template
<
>
void
matmul
<
platform
::
GPUPlace
,
float
>
(
const
framework
::
Tensor
&
matrix_a
,
bool
trans_a
,
const
framework
::
Tensor
&
matrix_b
,
bool
trans_b
,
float
alpha
,
framework
::
Tensor
*
matrix_out
,
float
beta
,
platform
::
DeviceContext
*
context
)
{
auto
dim_a
=
matrix_a
.
dims
();
auto
dim_b
=
matrix_b
.
dims
();
auto
dim_out
=
matrix_out
->
dims
();
PADDLE_ENFORCE
(
dim_a
.
size
()
==
2
&&
dim_b
.
size
()
==
2
&&
dim_out
.
size
()
==
2
,
"The input and output of matmul be matrix"
);
PADDLE_ENFORCE
(
platform
::
is_gpu_place
(
matrix_a
.
place
())
&&
platform
::
is_gpu_place
(
matrix_b
.
place
())
&&
platform
::
is_gpu_place
(
matrix_out
->
place
()),
"Matrix must all be in GPUPlace"
);
int
M
=
dim_out
[
0
];
int
N
=
dim_out
[
1
];
int
K
=
(
trans_a
==
false
)
?
dim_a
[
1
]
:
dim_a
[
0
];
CBLAS_TRANSPOSE
transA
=
(
trans_a
==
false
)
?
CblasNoTrans
:
CblasTrans
;
CBLAS_TRANSPOSE
transB
=
(
trans_b
==
false
)
?
CblasNoTrans
:
CblasTrans
;
gemm
<
platform
::
GPUPlace
,
float
>
(
transA
,
transB
,
M
,
N
,
K
,
alpha
,
matrix_a
.
data
<
float
>
(),
matrix_b
.
data
<
float
>
(),
beta
,
matrix_out
->
data
<
float
>
(),
context
);
}
template
<
>
void
matmul
<
platform
::
GPUPlace
,
double
>
(
const
framework
::
Tensor
&
matrix_a
,
bool
trans_a
,
const
framework
::
Tensor
&
matrix_b
,
bool
trans_b
,
double
alpha
,
framework
::
Tensor
*
matrix_out
,
double
beta
,
platform
::
DeviceContext
*
context
)
{
auto
dim_a
=
matrix_a
.
dims
();
auto
dim_b
=
matrix_b
.
dims
();
auto
dim_out
=
matrix_out
->
dims
();
PADDLE_ENFORCE
(
dim_a
.
size
()
==
2
&&
dim_b
.
size
()
==
2
&&
dim_out
.
size
()
==
2
,
"The input and output of matmul be matrix"
);
PADDLE_ENFORCE
(
platform
::
is_gpu_place
(
matrix_a
.
place
())
&&
platform
::
is_gpu_place
(
matrix_b
.
place
())
&&
platform
::
is_gpu_place
(
matrix_out
->
place
()),
"Matrix must all be in GPUPlace"
);
int
M
=
dim_out
[
0
];
int
N
=
dim_out
[
1
];
int
K
=
(
trans_a
==
false
)
?
dim_a
[
1
]
:
dim_a
[
0
];
CBLAS_TRANSPOSE
transA
=
(
trans_a
==
false
)
?
CblasNoTrans
:
CblasTrans
;
CBLAS_TRANSPOSE
transB
=
(
trans_b
==
false
)
?
CblasNoTrans
:
CblasTrans
;
gemm
<
platform
::
GPUPlace
,
double
>
(
transA
,
transB
,
M
,
N
,
K
,
alpha
,
matrix_a
.
data
<
double
>
(),
matrix_b
.
data
<
double
>
(),
beta
,
matrix_out
->
data
<
double
>
(),
context
);
}
}
// namespace math
}
// namespace operators
}
// namespace paddle
paddle/operators/math/math_function.h
0 → 100644
浏览文件 @
2f74e608
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
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
#ifdef PADDLE_USE_MKLML
#include <mkl_cblas.h>
#include <mkl_lapacke.h>
#include <mkl_vml_functions.h>
#endif
#ifdef PADDLE_USE_MKL
#include <mkl.h>
#include <mkl_lapacke.h>
#endif
#ifdef PADDLE_USE_ATLAS
extern
"C"
{
#include <cblas.h>
#include <clapack.h>
}
#endif
#ifdef PADDLE_USE_OPENBLAS
#include <cblas.h>
#include <lapacke.h>
#endif
#ifndef LAPACK_FOUND
extern
"C"
{
#include <cblas.h>
int
LAPACKE_sgetrf
(
int
matrix_layout
,
int
m
,
int
n
,
float
*
a
,
int
lda
,
int
*
ipiv
);
int
LAPACKE_dgetrf
(
int
matrix_layout
,
int
m
,
int
n
,
double
*
a
,
int
lda
,
int
*
ipiv
);
int
LAPACKE_sgetri
(
int
matrix_layout
,
int
n
,
float
*
a
,
int
lda
,
const
int
*
ipiv
);
int
LAPACKE_dgetri
(
int
matrix_layout
,
int
n
,
double
*
a
,
int
lda
,
const
int
*
ipiv
);
}
#endif
#include <cmath>
#include "paddle/framework/tensor.h"
#include "paddle/platform/device_context.h"
#include "paddle/platform/enforce.h"
namespace
paddle
{
namespace
operators
{
namespace
math
{
// Support continuous memory now
// If transA = N, and transB = N
// Then matrixA: M * K, matrixB: K * N matrixC : M * N
// For more detailed info, please refer to
// http://www.netlib.org/lapack/explore-html/d4/de2/sgemm_8f.html
template
<
typename
Place
,
typename
T
>
void
gemm
(
const
CBLAS_TRANSPOSE
transA
,
const
CBLAS_TRANSPOSE
transB
,
const
int
M
,
const
int
N
,
const
int
K
,
const
T
alpha
,
const
T
*
A
,
const
T
*
B
,
const
T
beta
,
T
*
C
,
platform
::
DeviceContext
*
context
);
// matrix multiply with continuous memory
template
<
typename
Place
,
typename
T
>
void
matmul
(
const
framework
::
Tensor
&
matrix_a
,
bool
trans_a
,
const
framework
::
Tensor
&
matrix_b
,
bool
trans_b
,
T
alpha
,
framework
::
Tensor
*
matrix_out
,
T
beta
,
platform
::
DeviceContext
*
context
);
}
// namespace math
}
// namespace operators
}
// namespace paddle
paddle/operators/math/math_function_test.cc
0 → 100644
浏览文件 @
2f74e608
#include "paddle/operators/math/math_function.h"
#include "gtest/gtest.h"
#ifndef PADDLE_ONLY_CPU
TEST
(
math_function
,
notrans_mul_trans
)
{
paddle
::
framework
::
Tensor
input1
;
paddle
::
framework
::
Tensor
input1_gpu
;
paddle
::
framework
::
Tensor
input2_gpu
;
paddle
::
framework
::
Tensor
out_gpu
;
paddle
::
framework
::
Tensor
out
;
auto
*
cpu_place
=
new
paddle
::
platform
::
CPUPlace
();
float
*
input1_ptr
=
input1
.
mutable_data
<
float
>
({
2
,
3
},
*
cpu_place
);
float
arr
[
6
]
=
{
0
,
1
,
2
,
3
,
4
,
5
};
memcpy
(
input1_ptr
,
arr
,
6
*
sizeof
(
float
));
auto
*
gpu_place
=
new
paddle
::
platform
::
GPUPlace
(
0
);
paddle
::
platform
::
DeviceContext
*
context
=
new
paddle
::
platform
::
CUDADeviceContext
(
*
gpu_place
);
input1_gpu
.
CopyFrom
<
float
>
(
input1
,
*
gpu_place
);
input2_gpu
.
CopyFrom
<
float
>
(
input1
,
*
gpu_place
);
out_gpu
.
mutable_data
<
float
>
({
2
,
2
},
*
gpu_place
);
paddle
::
operators
::
math
::
matmul
<
paddle
::
platform
::
GPUPlace
,
float
>
(
input1_gpu
,
false
,
input2_gpu
,
true
,
1
,
&
out_gpu
,
0
,
context
);
out
.
CopyFrom
<
float
>
(
out_gpu
,
*
cpu_place
);
float
*
out_ptr
=
out
.
data
<
float
>
();
EXPECT_EQ
(
out_ptr
[
0
],
5
);
EXPECT_EQ
(
out_ptr
[
1
],
14
);
EXPECT_EQ
(
out_ptr
[
2
],
14
);
EXPECT_EQ
(
out_ptr
[
3
],
50
);
}
TEST
(
math_function
,
trans_mul_notrans
)
{
paddle
::
framework
::
Tensor
input1
;
paddle
::
framework
::
Tensor
input1_gpu
;
paddle
::
framework
::
Tensor
input2_gpu
;
paddle
::
framework
::
Tensor
out_gpu
;
paddle
::
framework
::
Tensor
out
;
auto
*
cpu_place
=
new
paddle
::
platform
::
CPUPlace
();
float
*
input1_ptr
=
input1
.
mutable_data
<
float
>
({
2
,
3
},
*
cpu_place
);
float
arr
[
6
]
=
{
0
,
1
,
2
,
3
,
4
,
5
};
memcpy
(
input1_ptr
,
arr
,
6
*
sizeof
(
float
));
auto
*
gpu_place
=
new
paddle
::
platform
::
GPUPlace
(
0
);
paddle
::
platform
::
DeviceContext
*
context
=
new
paddle
::
platform
::
CUDADeviceContext
(
*
gpu_place
);
input1_gpu
.
CopyFrom
<
float
>
(
input1
,
*
gpu_place
);
input2_gpu
.
CopyFrom
<
float
>
(
input1
,
*
gpu_place
);
out_gpu
.
mutable_data
<
float
>
({
3
,
3
},
*
gpu_place
);
paddle
::
operators
::
math
::
matmul
<
paddle
::
platform
::
GPUPlace
,
float
>
(
input1_gpu
,
true
,
input2_gpu
,
false
,
1
,
&
out_gpu
,
0
,
context
);
out
.
CopyFrom
<
float
>
(
out_gpu
,
*
cpu_place
);
float
*
out_ptr
=
out
.
data
<
float
>
();
EXPECT_EQ
(
out_ptr
[
0
],
9
);
EXPECT_EQ
(
out_ptr
[
1
],
12
);
EXPECT_EQ
(
out_ptr
[
2
],
15
);
EXPECT_EQ
(
out_ptr
[
3
],
12
);
EXPECT_EQ
(
out_ptr
[
4
],
17
);
EXPECT_EQ
(
out_ptr
[
5
],
22
);
EXPECT_EQ
(
out_ptr
[
6
],
15
);
EXPECT_EQ
(
out_ptr
[
7
],
22
);
EXPECT_EQ
(
out_ptr
[
8
],
29
);
}
#endif
paddle/operators/mean_op.cc
浏览文件 @
2f74e608
...
@@ -18,7 +18,9 @@ namespace paddle {
...
@@ -18,7 +18,9 @@ namespace paddle {
namespace
operators
{
namespace
operators
{
class
MeanOp
:
public
framework
::
OperatorWithKernel
{
class
MeanOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
MeanOp
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
PADDLE_ENFORCE_NOT_NULL
(
ctx
.
InputVar
(
"X"
),
PADDLE_ENFORCE_NOT_NULL
(
ctx
.
InputVar
(
"X"
),
...
@@ -38,7 +40,9 @@ class MeanOpMaker : public framework::OpProtoAndCheckerMaker {
...
@@ -38,7 +40,9 @@ class MeanOpMaker : public framework::OpProtoAndCheckerMaker {
};
};
class
MeanGradOp
:
public
framework
::
OperatorWithKernel
{
class
MeanGradOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
MeanGradOp
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
ctx
.
Output
<
Tensor
>
(
framework
::
GradVarName
(
"X"
))
ctx
.
Output
<
Tensor
>
(
framework
::
GradVarName
(
"X"
))
...
...
paddle/operators/mul_op.cc
浏览文件 @
2f74e608
...
@@ -13,12 +13,14 @@
...
@@ -13,12 +13,14 @@
limitations under the License. */
limitations under the License. */
#include "paddle/operators/mul_op.h"
#include "paddle/operators/mul_op.h"
#include "paddle/operators/math/math_function.h"
namespace
paddle
{
namespace
paddle
{
namespace
operators
{
namespace
operators
{
class
MulOp
:
public
framework
::
OperatorWithKernel
{
class
MulOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
MulOp
,
framework
::
OperatorWithKernel
);
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
...
@@ -53,7 +55,9 @@ The equation is: Out = X * Y
...
@@ -53,7 +55,9 @@ The equation is: Out = X * Y
};
};
class
MulOpGrad
:
public
framework
::
OperatorWithKernel
{
class
MulOpGrad
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
MulOpGrad
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{}
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{}
std
::
string
DebugString
()
const
override
{
std
::
string
DebugString
()
const
override
{
...
...
paddle/operators/mul_op.cu
浏览文件 @
2f74e608
...
@@ -16,5 +16,4 @@
...
@@ -16,5 +16,4 @@
#include "paddle/operators/mul_op.h"
#include "paddle/operators/mul_op.h"
namespace
ops
=
paddle
::
operators
;
namespace
ops
=
paddle
::
operators
;
REGISTER_OP_GPU_KERNEL
(
mul
,
ops
::
MulKernel
<
paddle
::
platform
::
GPUPlace
,
float
>
);
REGISTER_OP_GPU_KERNEL
(
mul
,
ops
::
MulKernel
<
paddle
::
platform
::
GPUPlace
,
float
>
);
paddle/operators/mul_op.h
浏览文件 @
2f74e608
...
@@ -13,6 +13,9 @@
...
@@ -13,6 +13,9 @@
limitations under the License. */
limitations under the License. */
#pragma once
#pragma once
#include "paddle/operators/math/math_function.h"
#include "paddle/framework/eigen.h"
#include "paddle/framework/eigen.h"
#include "paddle/framework/op_registry.h"
#include "paddle/framework/op_registry.h"
...
...
paddle/operators/net_op.cc
浏览文件 @
2f74e608
...
@@ -81,5 +81,11 @@ std::vector<std::string> NetOp::OutputVars(bool has_intermediate) const {
...
@@ -81,5 +81,11 @@ std::vector<std::string> NetOp::OutputVars(bool has_intermediate) const {
return
ret_val
;
return
ret_val
;
}
}
NetOp
::
NetOp
(
const
std
::
string
&
type
,
const
framework
::
OperatorBase
::
VarNameMap
&
inputs
,
const
framework
::
OperatorBase
::
VarNameMap
&
outputs
,
const
framework
::
AttributeMap
&
attrs
)
:
OperatorBase
(
type
,
inputs
,
outputs
,
attrs
)
{}
}
// namespace operators
}
// namespace operators
}
// namespace paddle
}
// namespace paddle
paddle/operators/net_op.h
浏览文件 @
2f74e608
...
@@ -37,7 +37,9 @@ namespace operators {
...
@@ -37,7 +37,9 @@ namespace operators {
class
NetOp
:
public
framework
::
OperatorBase
{
class
NetOp
:
public
framework
::
OperatorBase
{
public:
public:
static
const
char
kAll
[];
static
const
char
kAll
[];
DEFINE_OPERATOR_CTOR
(
NetOp
,
framework
::
OperatorBase
);
NetOp
()
:
framework
::
OperatorBase
(
"plain_net"
,
{},
{},
{})
{}
NetOp
(
const
std
::
string
&
type
,
const
VarNameMap
&
inputs
,
const
VarNameMap
&
outputs
,
const
framework
::
AttributeMap
&
attrs
);
/**
/**
* Infer all the operators' input and output variables' shapes, will be called
* Infer all the operators' input and output variables' shapes, will be called
...
...
paddle/operators/net_op_test.cc
浏览文件 @
2f74e608
...
@@ -12,7 +12,7 @@ static int run_cnt = 0;
...
@@ -12,7 +12,7 @@ static int run_cnt = 0;
class
TestOp
:
public
framework
::
OperatorBase
{
class
TestOp
:
public
framework
::
OperatorBase
{
public:
public:
DEFINE_OPERATOR_CTOR
(
TestOp
,
framework
::
OperatorBase
)
;
using
framework
::
OperatorBase
::
OperatorBase
;
void
InferShape
(
const
Scope
&
scope
)
const
override
{
++
infer_shape_cnt
;
}
void
InferShape
(
const
Scope
&
scope
)
const
override
{
++
infer_shape_cnt
;
}
void
Run
(
const
Scope
&
scope
,
void
Run
(
const
Scope
&
scope
,
const
platform
::
DeviceContext
&
dev_ctx
)
const
override
{
const
platform
::
DeviceContext
&
dev_ctx
)
const
override
{
...
@@ -22,7 +22,7 @@ class TestOp : public framework::OperatorBase {
...
@@ -22,7 +22,7 @@ class TestOp : public framework::OperatorBase {
class
EmptyOp
:
public
framework
::
OperatorBase
{
class
EmptyOp
:
public
framework
::
OperatorBase
{
public:
public:
DEFINE_OPERATOR_CTOR
(
EmptyOp
,
framework
::
OperatorBase
)
;
using
framework
::
OperatorBase
::
OperatorBase
;
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
void
InferShape
(
const
Scope
&
scope
)
const
override
{}
void
Run
(
const
Scope
&
scope
,
const
DeviceContext
&
dev_ctx
)
const
override
{}
void
Run
(
const
Scope
&
scope
,
const
DeviceContext
&
dev_ctx
)
const
override
{}
};
};
...
@@ -44,14 +44,14 @@ TEST(OpKernel, all) {
...
@@ -44,14 +44,14 @@ TEST(OpKernel, all) {
auto
net
=
std
::
make_shared
<
NetOp
>
();
auto
net
=
std
::
make_shared
<
NetOp
>
();
ASSERT_NE
(
net
,
nullptr
);
ASSERT_NE
(
net
,
nullptr
);
auto
op1
=
std
::
make_shared
<
TestOp
>
();
auto
op1
=
std
::
shared_ptr
<
TestOp
>
(
op1
->
inputs_
=
{{
"X"
,
{
"x"
}},
{
"W"
,
{
"w1"
}},
{
"b"
,
{
"b1"
}}};
new
TestOp
(
"test"
,
{{
"X"
,
{
"x"
}},
{
"W"
,
{
"w1"
}},
{
"b"
,
{
"b1"
}}},
op1
->
outputs_
=
{{
"Out"
,
{
"y"
}}}
;
{{
"Out"
,
{
"y"
}}},
{}))
;
net
->
AddOp
(
op1
);
net
->
AddOp
(
op1
);
auto
op2
=
std
::
make_shared
<
TestOp
>
();
auto
op2
=
std
::
shared_ptr
<
TestOp
>
(
op2
->
inputs_
=
{{
"X"
,
{
"y"
}},
{
"W"
,
{
"w2"
}},
{
"b"
,
{
"b2"
}}};
new
TestOp
(
"test"
,
{{
"X"
,
{
"y"
}},
{
"W"
,
{
"w2"
}},
{
"b"
,
{
"b2"
}}},
op2
->
outputs_
=
{{
"Out"
,
{
"z"
}}}
;
{{
"Out"
,
{
"z"
}}},
{}))
;
net
->
AddOp
(
op2
);
net
->
AddOp
(
op2
);
net
->
CompleteAddOp
();
net
->
CompleteAddOp
();
...
@@ -67,9 +67,9 @@ TEST(OpKernel, all) {
...
@@ -67,9 +67,9 @@ TEST(OpKernel, all) {
TEST
(
NetOp
,
insert_op
)
{
TEST
(
NetOp
,
insert_op
)
{
NetOp
net
;
NetOp
net
;
auto
op1
=
std
::
make_shared
<
EmptyOp
>
();
auto
op1
=
std
::
shared_ptr
<
EmptyOp
>
(
op1
->
inputs_
=
{{
"X"
,
{
"x"
}},
{
"W"
,
{
"w1"
}},
{
"b"
,
{
"b1"
}}};
new
EmptyOp
(
"empty"
,
{{
"X"
,
{
"x"
}},
{
"W"
,
{
"w1"
}},
{
"b"
,
{
"b1"
}}},
op1
->
outputs_
=
{{
"Out"
,
{
"y"
}}}
;
{{
"Out"
,
{
"y"
}}},
{}))
;
net
.
AddOp
(
op1
);
net
.
AddOp
(
op1
);
net
.
InsertOp
(
0
,
op1
);
net
.
InsertOp
(
0
,
op1
);
ASSERT_EQ
(
2UL
,
net
.
ops_
.
size
());
ASSERT_EQ
(
2UL
,
net
.
ops_
.
size
());
...
...
paddle/operators/recurrent_op.cc
浏览文件 @
2f74e608
...
@@ -135,8 +135,11 @@ const rnn::ArgumentName RecurrentGradientOp::kArgName{
...
@@ -135,8 +135,11 @@ const rnn::ArgumentName RecurrentGradientOp::kArgName{
"inlink@grad"
,
"inlink_alias"
,
"outlink_alias"
,
"inlink@grad"
,
"inlink_alias"
,
"outlink_alias"
,
"memories"
,
"pre_memories"
,
"boot_memories@grad"
};
"memories"
,
"pre_memories"
,
"boot_memories@grad"
};
void
RecurrentOp
::
Init
()
{
RecurrentOp
::
RecurrentOp
(
const
std
::
string
&
type
,
OperatorBase
::
Init
();
const
framework
::
OperatorBase
::
VarNameMap
&
inputs
,
const
framework
::
OperatorBase
::
VarNameMap
&
outputs
,
const
framework
::
AttributeMap
&
attrs
)
:
OperatorBase
(
type
,
inputs
,
outputs
,
attrs
)
{
std
::
unique_ptr
<
rnn
::
Argument
>
arg
(
new
rnn
::
Argument
());
std
::
unique_ptr
<
rnn
::
Argument
>
arg
(
new
rnn
::
Argument
());
rnn
::
InitArgument
(
kArgName
,
arg
.
get
(),
*
this
);
rnn
::
InitArgument
(
kArgName
,
arg
.
get
(),
*
this
);
alg_
.
Init
(
std
::
move
(
arg
));
alg_
.
Init
(
std
::
move
(
arg
));
...
@@ -230,8 +233,11 @@ void RecurrentGradientAlgorithm::InferShape(const Scope& scope) const {
...
@@ -230,8 +233,11 @@ void RecurrentGradientAlgorithm::InferShape(const Scope& scope) const {
LinkBootMemoryGradients
(
step_scopes
[
0
],
true
/*infer_shape_mode*/
);
LinkBootMemoryGradients
(
step_scopes
[
0
],
true
/*infer_shape_mode*/
);
}
}
void
RecurrentGradientOp
::
Init
()
{
RecurrentGradientOp
::
RecurrentGradientOp
(
OperatorBase
::
Init
();
const
std
::
string
&
type
,
const
framework
::
OperatorBase
::
VarNameMap
&
inputs
,
const
framework
::
OperatorBase
::
VarNameMap
&
outputs
,
const
framework
::
AttributeMap
&
attrs
)
:
OperatorBase
(
type
,
inputs
,
outputs
,
attrs
)
{
std
::
unique_ptr
<
rnn
::
Argument
>
arg
(
new
rnn
::
Argument
());
std
::
unique_ptr
<
rnn
::
Argument
>
arg
(
new
rnn
::
Argument
());
rnn
::
InitArgument
(
kArgName
,
arg
.
get
(),
*
this
);
rnn
::
InitArgument
(
kArgName
,
arg
.
get
(),
*
this
);
alg_
.
Init
(
std
::
move
(
arg
));
alg_
.
Init
(
std
::
move
(
arg
));
...
...
paddle/operators/recurrent_op.h
浏览文件 @
2f74e608
...
@@ -101,13 +101,11 @@ class RecurrentGradientAlgorithm {
...
@@ -101,13 +101,11 @@ class RecurrentGradientAlgorithm {
class
RecurrentOp
final
:
public
framework
::
OperatorBase
{
class
RecurrentOp
final
:
public
framework
::
OperatorBase
{
public:
public:
DEFINE_OPERATOR_CTOR
(
RecurrentOp
,
framework
::
OperatorBase
);
RecurrentOp
(
const
std
::
string
&
type
,
const
VarNameMap
&
inputs
,
const
VarNameMap
&
outputs
,
const
framework
::
AttributeMap
&
attrs
);
void
Init
()
override
;
/**
/**
* InferShape must be called before Run.
* InferShape must be called before Run.
*/
*/
void
InferShape
(
const
framework
::
Scope
&
scope
)
const
override
{
void
InferShape
(
const
framework
::
Scope
&
scope
)
const
override
{
alg_
.
InferShape
(
scope
);
alg_
.
InferShape
(
scope
);
}
}
...
@@ -125,8 +123,9 @@ class RecurrentOp final : public framework::OperatorBase {
...
@@ -125,8 +123,9 @@ class RecurrentOp final : public framework::OperatorBase {
class
RecurrentGradientOp
final
:
public
framework
::
OperatorBase
{
class
RecurrentGradientOp
final
:
public
framework
::
OperatorBase
{
public:
public:
DEFINE_OPERATOR_CTOR
(
RecurrentGradientOp
,
framework
::
OperatorBase
)
RecurrentGradientOp
(
const
std
::
string
&
type
,
const
VarNameMap
&
inputs
,
void
Init
()
override
;
const
VarNameMap
&
outputs
,
const
framework
::
AttributeMap
&
attrs
);
/**
/**
* InferShape must be called before Run.
* InferShape must be called before Run.
...
...
paddle/operators/rowwise_add_op.cc
浏览文件 @
2f74e608
...
@@ -18,7 +18,9 @@ namespace paddle {
...
@@ -18,7 +18,9 @@ namespace paddle {
namespace
operators
{
namespace
operators
{
class
RowWiseAddOp
:
public
framework
::
OperatorWithKernel
{
class
RowWiseAddOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
RowWiseAddOp
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
auto
dim0
=
ctx
.
Input
<
Tensor
>
(
"X"
)
->
dims
();
auto
dim0
=
ctx
.
Input
<
Tensor
>
(
"X"
)
->
dims
();
...
...
paddle/operators/sgd_op.cc
浏览文件 @
2f74e608
...
@@ -18,7 +18,9 @@ namespace paddle {
...
@@ -18,7 +18,9 @@ namespace paddle {
namespace
operators
{
namespace
operators
{
class
SGDOp
:
public
framework
::
OperatorWithKernel
{
class
SGDOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
SGDOp
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
PADDLE_ENFORCE
(
PADDLE_ENFORCE
(
...
...
paddle/operators/sigmoid_op.cc
浏览文件 @
2f74e608
...
@@ -18,7 +18,9 @@ namespace paddle {
...
@@ -18,7 +18,9 @@ namespace paddle {
namespace
operators
{
namespace
operators
{
class
SigmoidOp
:
public
framework
::
OperatorWithKernel
{
class
SigmoidOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
SigmoidOp
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
ctx
.
Output
<
Tensor
>
(
"Y"
)
->
Resize
(
ctx
.
Input
<
Tensor
>
(
"X"
)
->
dims
());
ctx
.
Output
<
Tensor
>
(
"Y"
)
->
Resize
(
ctx
.
Input
<
Tensor
>
(
"X"
)
->
dims
());
...
@@ -37,7 +39,9 @@ class SigmoidOpMaker : public framework::OpProtoAndCheckerMaker {
...
@@ -37,7 +39,9 @@ class SigmoidOpMaker : public framework::OpProtoAndCheckerMaker {
};
};
class
SigmoidOpGrad
:
public
framework
::
OperatorWithKernel
{
class
SigmoidOpGrad
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
SigmoidOpGrad
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
ctx
.
Output
<
Tensor
>
(
0
)
->
Resize
(
ctx
.
Input
<
Tensor
>
(
0
)
->
dims
());
ctx
.
Output
<
Tensor
>
(
0
)
->
Resize
(
ctx
.
Input
<
Tensor
>
(
0
)
->
dims
());
...
...
paddle/operators/softmax_op.cc
浏览文件 @
2f74e608
...
@@ -18,7 +18,9 @@ namespace paddle {
...
@@ -18,7 +18,9 @@ namespace paddle {
namespace
operators
{
namespace
operators
{
class
SoftmaxOp
:
public
framework
::
OperatorWithKernel
{
class
SoftmaxOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
SoftmaxOp
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
PADDLE_ENFORCE
(
ctx
.
Input
<
Tensor
>
(
"X"
)
->
dims
().
size
()
==
2UL
,
PADDLE_ENFORCE
(
ctx
.
Input
<
Tensor
>
(
"X"
)
->
dims
().
size
()
==
2UL
,
...
@@ -39,7 +41,9 @@ class SoftmaxOpMaker : public framework::OpProtoAndCheckerMaker {
...
@@ -39,7 +41,9 @@ class SoftmaxOpMaker : public framework::OpProtoAndCheckerMaker {
};
};
class
SoftmaxOpGrad
:
public
framework
::
OperatorWithKernel
{
class
SoftmaxOpGrad
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
SoftmaxOpGrad
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
PADDLE_ENFORCE
(
ctx
.
InputVar
(
"Y"
)
!=
nullptr
,
"Input(Y) should not be null"
);
PADDLE_ENFORCE
(
ctx
.
InputVar
(
"Y"
)
!=
nullptr
,
"Input(Y) should not be null"
);
...
...
paddle/operators/uniform_random_op.cc
浏览文件 @
2f74e608
...
@@ -46,7 +46,9 @@ class CPUUniformRandomKernel : public framework::OpKernel {
...
@@ -46,7 +46,9 @@ class CPUUniformRandomKernel : public framework::OpKernel {
};
};
class
UniformRandomOp
:
public
framework
::
OperatorWithKernel
{
class
UniformRandomOp
:
public
framework
::
OperatorWithKernel
{
DEFINE_OPERATOR_CTOR
(
UniformRandomOp
,
framework
::
OperatorWithKernel
)
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
PADDLE_ENFORCE
(
GetAttr
<
float
>
(
"min"
)
<
GetAttr
<
float
>
(
"max"
),
PADDLE_ENFORCE
(
GetAttr
<
float
>
(
"min"
)
<
GetAttr
<
float
>
(
"max"
),
...
...
paddle/platform/dynload/cublas.h
浏览文件 @
2f74e608
...
@@ -62,12 +62,12 @@ extern void *cublas_dso_handle;
...
@@ -62,12 +62,12 @@ extern void *cublas_dso_handle;
DECLARE_DYNAMIC_LOAD_CUBLAS_WRAP(__name)
DECLARE_DYNAMIC_LOAD_CUBLAS_WRAP(__name)
#define CUBLAS_BLAS_ROUTINE_EACH(__macro) \
#define CUBLAS_BLAS_ROUTINE_EACH(__macro) \
__macro(cublasSgemv
);
\
__macro(cublasSgemv
_v2);
\
__macro(cublasDgemv
);
\
__macro(cublasDgemv
_v2);
\
__macro(cublasSgemm
);
\
__macro(cublasSgemm
_v2);
\
__macro(cublasDgemm
);
\
__macro(cublasDgemm
_v2);
\
__macro(cublasSgeam
);
\
__macro(cublasSgeam
_v2);
\
__macro(cublasDgeam
);
\
__macro(cublasDgeam
_v2);
\
__macro(cublasCreate_v2); \
__macro(cublasCreate_v2); \
__macro(cublasDestroy_v2); \
__macro(cublasDestroy_v2); \
__macro(cublasSetStream_v2); \
__macro(cublasSetStream_v2); \
...
...
paddle/platform/enforce.h
浏览文件 @
2f74e608
...
@@ -14,14 +14,21 @@ limitations under the License. */
...
@@ -14,14 +14,21 @@ limitations under the License. */
#pragma once
#pragma once
#include <execinfo.h>
#include <dlfcn.h> // for dladdr
#include <execinfo.h> // for backtrace
#include <iomanip>
#include <iomanip>
#include <memory>
#include <sstream>
#include <sstream>
#include <stdexcept>
#include <stdexcept>
#include <string>
#include <string>
#include "paddle/string/printf.h"
#include "paddle/string/printf.h"
#include "paddle/string/to_string.h"
#include "paddle/string/to_string.h"
#ifdef __GNUC__
#include <cxxabi.h> // for __cxa_demangle
#endif
#ifndef PADDLE_ONLY_CPU
#ifndef PADDLE_ONLY_CPU
#include "paddle/platform/dynload/cublas.h"
#include "paddle/platform/dynload/cublas.h"
...
@@ -39,6 +46,19 @@ limitations under the License. */
...
@@ -39,6 +46,19 @@ limitations under the License. */
namespace
paddle
{
namespace
paddle
{
namespace
platform
{
namespace
platform
{
namespace
{
#ifdef __GNUC__
inline
std
::
string
demangle
(
std
::
string
name
)
{
int
status
=
-
4
;
// some arbitrary value to eliminate the compiler warning
std
::
unique_ptr
<
char
,
void
(
*
)(
void
*
)
>
res
{
abi
::
__cxa_demangle
(
name
.
c_str
(),
NULL
,
NULL
,
&
status
),
std
::
free
};
return
(
status
==
0
)
?
res
.
get
()
:
name
;
}
#else
inline
std
::
string
demangle
(
std
::
string
name
)
{
return
name
;
}
#endif
}
struct
EnforceNotMet
:
public
std
::
exception
{
struct
EnforceNotMet
:
public
std
::
exception
{
std
::
exception_ptr
exp_
;
std
::
exception_ptr
exp_
;
std
::
string
err_str_
;
std
::
string
err_str_
;
...
@@ -48,15 +68,29 @@ struct EnforceNotMet : public std::exception {
...
@@ -48,15 +68,29 @@ struct EnforceNotMet : public std::exception {
std
::
rethrow_exception
(
exp_
);
std
::
rethrow_exception
(
exp_
);
}
catch
(
const
std
::
exception
&
exp
)
{
}
catch
(
const
std
::
exception
&
exp
)
{
std
::
ostringstream
sout
;
std
::
ostringstream
sout
;
sout
<<
string
::
Sprintf
(
"%s at [%s:%d]"
,
exp
.
what
(),
f
,
l
)
<<
std
::
endl
;
sout
<<
string
::
Sprintf
(
"%s at [%s:%d]"
,
exp
.
what
(),
f
,
l
)
<<
std
::
endl
;
sout
<<
"Call Stacks: "
<<
std
::
endl
;
sout
<<
"PaddlePaddle Call Stacks: "
<<
std
::
endl
;
void
*
call_stack
[
TRACE_STACK_LIMIT
];
void
*
call_stack
[
TRACE_STACK_LIMIT
];
int
sz
=
backtrace
(
call_stack
,
TRACE_STACK_LIMIT
);
auto
size
=
backtrace
(
call_stack
,
TRACE_STACK_LIMIT
);
auto
line
=
backtrace_symbols
(
call_stack
,
sz
);
auto
symbols
=
backtrace_symbols
(
call_stack
,
size
);
for
(
int
i
=
0
;
i
<
sz
;
++
i
)
{
sout
<<
line
[
i
]
<<
std
::
endl
;
Dl_info
info
;
for
(
int
i
=
0
;
i
<
size
;
++
i
)
{
if
(
dladdr
(
call_stack
[
i
],
&
info
))
{
auto
demangled
=
demangle
(
info
.
dli_sname
);
auto
addr_offset
=
static_cast
<
char
*>
(
call_stack
[
i
])
-
static_cast
<
char
*>
(
info
.
dli_saddr
);
sout
<<
string
::
Sprintf
(
"%-3d %*0p %s + %zd
\n
"
,
i
,
2
+
sizeof
(
void
*
)
*
2
,
call_stack
[
i
],
demangled
,
addr_offset
);
}
else
{
sout
<<
string
::
Sprintf
(
"%-3d %*0p %s
\n
"
,
i
,
2
+
sizeof
(
void
*
)
*
2
,
call_stack
[
i
]);
}
}
}
free
(
line
);
free
(
symbols
);
err_str_
=
sout
.
str
();
err_str_
=
sout
.
str
();
}
}
}
}
...
@@ -170,7 +204,7 @@ inline void throw_on_error(T e) {
...
@@ -170,7 +204,7 @@ inline void throw_on_error(T e) {
* PADDLE_ENFORCE_EQ(a, b);
* PADDLE_ENFORCE_EQ(a, b);
*
*
* will raise an expression described as follows:
* will raise an expression described as follows:
* "enforce a == b failed, 1 != 2" with detailed stack infomation.
* "enforce a == b failed, 1 != 2" with detailed stack info
r
mation.
*
*
* extra messages is also supported, for example:
* extra messages is also supported, for example:
* PADDLE_ENFORCE(a, b, "some simple enforce failed between %d numbers", 2)
* PADDLE_ENFORCE(a, b, "some simple enforce failed between %d numbers", 2)
...
...
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