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
booniebears
Paddle
提交
a2a69f2a
P
Paddle
项目概览
booniebears
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
0
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 >>
提交
a2a69f2a
编写于
9月 07, 2017
作者:
Q
qingqing01
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Add function to get element count from tensor.
上级
2f40da09
变更
13
隐藏空白更改
内联
并排
Showing
13 changed file
with
33 addition
and
28 deletion
+33
-28
paddle/framework/tensor.h
paddle/framework/tensor.h
+6
-0
paddle/framework/tensor_impl.h
paddle/framework/tensor_impl.h
+8
-5
paddle/operators/cos_sim_op.h
paddle/operators/cos_sim_op.h
+2
-2
paddle/operators/gaussian_random_op.cc
paddle/operators/gaussian_random_op.cc
+1
-1
paddle/operators/gaussian_random_op.cu
paddle/operators/gaussian_random_op.cu
+2
-2
paddle/operators/lookup_table_op.cu
paddle/operators/lookup_table_op.cu
+2
-2
paddle/operators/lookup_table_op.h
paddle/operators/lookup_table_op.h
+2
-2
paddle/operators/mean_op.h
paddle/operators/mean_op.h
+2
-3
paddle/operators/minus_op.cc
paddle/operators/minus_op.cc
+1
-2
paddle/operators/squared_l2_distance_op.cc
paddle/operators/squared_l2_distance_op.cc
+2
-4
paddle/operators/squared_l2_distance_op.h
paddle/operators/squared_l2_distance_op.h
+2
-2
paddle/operators/uniform_random_op.cc
paddle/operators/uniform_random_op.cc
+1
-1
paddle/operators/uniform_random_op.cu
paddle/operators/uniform_random_op.cu
+2
-2
未找到文件。
paddle/framework/tensor.h
浏览文件 @
a2a69f2a
...
...
@@ -78,6 +78,9 @@ class Tensor {
/*! Return the dimensions of the memory block. */
inline
const
DDim
&
dims
()
const
;
/*! Return the numel of the memory block. */
inline
int64_t
numel
()
const
;
/*! Resize the dimensions of the memory block. */
inline
Tensor
&
Resize
(
const
DDim
&
dims
);
...
...
@@ -159,6 +162,9 @@ class Tensor {
/*! points to dimensions of memory block. */
DDim
dims_
;
/*! the element count of tensor. */
int64_t
numel_
;
/**
* @brief A PlaceHolder may be shared by more than one tensor.
*
...
...
paddle/framework/tensor_impl.h
浏览文件 @
a2a69f2a
...
...
@@ -24,7 +24,7 @@ inline void Tensor::check_memory_size() const {
PADDLE_ENFORCE_NOT_NULL
(
holder_
,
"Tenosr holds no memory. Call Tensor::mutable_data first."
);
PADDLE_ENFORCE_GE
(
holder_
->
size
(),
product
(
dims_
)
*
sizeof
(
T
)
+
offset_
,
holder_
->
size
(),
numel_
*
sizeof
(
T
)
+
offset_
,
"Tensor's dims_ is out of bound. Call Tensor::mutable_data "
"first to re-allocate memory.
\n
"
"or maybe the required data-type mismatches the data already stored."
);
...
...
@@ -54,11 +54,11 @@ inline T* Tensor::mutable_data(DDim dims, platform::Place place) {
template
<
typename
T
>
inline
T
*
Tensor
::
mutable_data
(
platform
::
Place
place
)
{
static_assert
(
std
::
is_pod
<
T
>::
value
,
"T must be POD"
);
PADDLE_ENFORCE_GT
(
product
(
dims_
)
,
0
,
PADDLE_ENFORCE_GT
(
numel_
,
0
,
"Tensor's numel must be larger than zero to call "
"Tensor::mutable_data. Call Tensor::set_dim first."
);
/* some versions of boost::variant don't have operator!= */
int64_t
size
=
product
(
dims_
)
*
sizeof
(
T
);
int64_t
size
=
numel_
*
sizeof
(
T
);
if
(
holder_
==
nullptr
||
!
(
holder_
->
place
()
==
place
)
||
holder_
->
size
()
<
size
+
offset_
)
{
if
(
platform
::
is_cpu_place
(
place
))
{
...
...
@@ -97,7 +97,7 @@ inline void Tensor::CopyFrom(const Tensor& src,
auto
dst_ptr
=
static_cast
<
void
*>
(
mutable_data
<
T
>
(
dst_place
));
auto
size
=
product
(
src
.
dims_
)
*
sizeof
(
T
);
auto
size
=
src
.
numel
(
)
*
sizeof
(
T
);
if
(
platform
::
is_cpu_place
(
src_place
)
&&
platform
::
is_cpu_place
(
dst_place
))
{
memory
::
Copy
(
boost
::
get
<
platform
::
CPUPlace
>
(
dst_place
),
dst_ptr
,
...
...
@@ -131,7 +131,7 @@ inline Tensor Tensor::Slice(const int& begin_idx, const int& end_idx) const {
PADDLE_ENFORCE_LT
(
begin_idx
,
end_idx
,
"Begin index must be less than end index."
);
PADDLE_ENFORCE_NE
(
dims_
[
0
],
1
,
"Can not slice a tensor with dims_[0] = 1."
);
size_t
base
=
product
(
dims_
)
/
dims_
[
0
];
size_t
base
=
numel_
/
dims_
[
0
];
Tensor
dst
;
dst
.
holder_
=
holder_
;
DDim
dst_dims
=
dims_
;
...
...
@@ -143,10 +143,13 @@ inline Tensor Tensor::Slice(const int& begin_idx, const int& end_idx) const {
inline
Tensor
&
Tensor
::
Resize
(
const
DDim
&
dims
)
{
dims_
=
dims
;
numel_
=
product
(
dims_
);
return
*
this
;
}
inline
const
DDim
&
Tensor
::
dims
()
const
{
return
dims_
;
}
inline
int64_t
Tensor
::
numel
()
const
{
return
numel_
;
}
}
// namespace framework
}
// namespace paddle
paddle/operators/cos_sim_op.h
浏览文件 @
a2a69f2a
...
...
@@ -42,7 +42,7 @@ class CosSimKernel : public framework::OpKernel {
output_y_norm
->
mutable_data
<
T
>
(
context
.
GetPlace
());
auto
dims
=
input_x
->
dims
();
int
size
=
static_cast
<
int
>
(
framework
::
product
(
dims
)
);
int
64_t
size
=
input_x
->
numel
(
);
auto
new_dims
=
framework
::
make_ddim
({
dims
[
0
],
size
/
dims
[
0
]});
auto
x
=
EigenMatrix
<
T
>::
From
(
*
input_x
,
new_dims
);
auto
y
=
EigenMatrix
<
T
>::
From
(
*
input_y
,
new_dims
);
...
...
@@ -72,7 +72,7 @@ class CosSimGradKernel : public framework::OpKernel {
auto
*
input_grad_z
=
context
.
Input
<
Tensor
>
(
framework
::
GradVarName
(
"Out"
));
auto
dims
=
input_x
->
dims
();
int
size
=
static_cast
<
int
>
(
framework
::
product
(
dims
)
);
int
64_t
size
=
input_x
->
numel
(
);
auto
new_dims
=
framework
::
make_ddim
({
dims
[
0
],
size
/
dims
[
0
]});
auto
x
=
EigenMatrix
<
T
>::
From
(
*
input_x
,
new_dims
);
auto
y
=
EigenMatrix
<
T
>::
From
(
*
input_y
,
new_dims
);
...
...
paddle/operators/gaussian_random_op.cc
浏览文件 @
a2a69f2a
...
...
@@ -31,7 +31,7 @@ class CPUGaussianRandomKernel : public framework::OpKernel {
}
engine
.
seed
(
seed
);
std
::
normal_distribution
<
T
>
dist
(
mean
,
std
);
int64_t
size
=
framework
::
product
(
tensor
->
dims
()
);
int64_t
size
=
tensor
->
numel
(
);
for
(
int64_t
i
=
0
;
i
<
size
;
++
i
)
{
data
[
i
]
=
dist
(
engine
);
}
...
...
paddle/operators/gaussian_random_op.cu
浏览文件 @
a2a69f2a
...
...
@@ -50,8 +50,8 @@ class GPUGaussianRandomKernel : public framework::OpKernel {
T
mean
=
static_cast
<
T
>
(
context
.
Attr
<
float
>
(
"mean"
));
T
std
=
static_cast
<
T
>
(
context
.
Attr
<
float
>
(
"std"
));
thrust
::
counting_iterator
<
unsigned
int
>
index_sequence_begin
(
0
);
ssize_t
N
=
framework
::
product
(
tensor
->
dims
()
);
thrust
::
transform
(
index_sequence_begin
,
index_sequence_begin
+
N
,
int64_t
size
=
tensor
->
numel
(
);
thrust
::
transform
(
index_sequence_begin
,
index_sequence_begin
+
size
,
thrust
::
device_ptr
<
T
>
(
data
),
GaussianGenerator
<
T
>
(
mean
,
std
,
seed
));
}
...
...
paddle/operators/lookup_table_op.cu
浏览文件 @
a2a69f2a
...
...
@@ -70,7 +70,7 @@ class LookupTableCUDAKernel : public framework::OpKernel {
size_t
N
=
table_t
->
dims
()[
0
];
size_t
D
=
table_t
->
dims
()[
1
];
size_t
K
=
product
(
ids_t
->
dims
()
);
size_t
K
=
ids_t
->
numel
(
);
auto
ids
=
ids_t
->
data
<
int32_t
>
();
auto
table
=
table_t
->
data
<
T
>
();
auto
output
=
output_t
->
mutable_data
<
T
>
(
context
.
GetPlace
());
...
...
@@ -91,7 +91,7 @@ class LookupTableGradCUDAKernel : public framework::OpKernel {
int
N
=
d_table_t
->
dims
()[
0
];
int
D
=
d_table_t
->
dims
()[
1
];
int
K
=
product
(
ids_t
->
dims
()
);
int
K
=
ids_t
->
numel
(
);
const
int32_t
*
ids
=
ids_t
->
data
<
int32_t
>
();
const
T
*
d_output
=
d_output_t
->
data
<
T
>
();
T
*
d_table
=
d_table_t
->
mutable_data
<
T
>
(
context
.
GetPlace
());
...
...
paddle/operators/lookup_table_op.h
浏览文件 @
a2a69f2a
...
...
@@ -35,7 +35,7 @@ class LookupTableKernel : public framework::OpKernel {
auto
ids
=
ids_t
->
data
<
int32_t
>
();
auto
table
=
table_t
->
data
<
T
>
();
auto
output
=
output_t
->
mutable_data
<
T
>
(
context
.
GetPlace
());
for
(
ssize_t
i
=
0
;
i
<
product
(
ids_t
->
dims
()
);
++
i
)
{
for
(
int64_t
i
=
0
;
i
<
ids_t
->
numel
(
);
++
i
)
{
PADDLE_ENFORCE_LT
(
ids
[
i
],
N
);
PADDLE_ENFORCE_GE
(
ids
[
i
],
0
);
memcpy
(
output
+
i
*
D
,
table
+
ids
[
i
]
*
D
,
D
*
sizeof
(
T
));
...
...
@@ -61,7 +61,7 @@ class LookupTableGradKernel : public framework::OpKernel {
t
.
device
(
context
.
GetEigenDevice
<
platform
::
CPUPlace
>
())
=
t
.
constant
(
static_cast
<
T
>
(
0
));
for
(
ssize_t
i
=
0
;
i
<
product
(
ids_t
->
dims
()
);
++
i
)
{
for
(
int64_t
i
=
0
;
i
<
ids_t
->
numel
(
);
++
i
)
{
PADDLE_ENFORCE_LT
(
ids
[
i
],
N
);
PADDLE_ENFORCE_GE
(
ids
[
i
],
0
);
for
(
int
j
=
0
;
j
<
D
;
++
j
)
{
...
...
paddle/operators/mean_op.h
浏览文件 @
a2a69f2a
...
...
@@ -49,12 +49,11 @@ class MeanGradKernel : public framework::OpKernel {
public:
void
Compute
(
const
framework
::
ExecutionContext
&
context
)
const
override
{
auto
OG
=
context
.
Input
<
Tensor
>
(
framework
::
GradVarName
(
"Out"
));
PADDLE_ENFORCE
(
framework
::
product
(
OG
->
dims
())
==
1
,
"Mean Gradient should be scalar"
);
PADDLE_ENFORCE
(
OG
->
numel
()
==
1
,
"Mean Gradient should be scalar"
);
auto
IG
=
context
.
Output
<
Tensor
>
(
framework
::
GradVarName
(
"X"
));
IG
->
mutable_data
<
T
>
(
context
.
GetPlace
());
T
ig_size
=
(
T
)
framework
::
product
(
IG
->
dims
());
T
ig_size
=
static_cast
<
T
>
(
IG
->
numel
());
Eigen
::
DSizes
<
int
,
1
>
bcast
(
ig_size
);
EigenVector
<
T
>::
Flatten
(
*
IG
).
device
(
context
.
GetEigenDevice
<
Place
>
())
=
...
...
paddle/operators/minus_op.cc
浏览文件 @
a2a69f2a
...
...
@@ -31,8 +31,7 @@ class MinusOp : public framework::OperatorWithKernel {
auto
*
right_tensor
=
ctx
.
Input
<
framework
::
Tensor
>
(
"Y"
);
PADDLE_ENFORCE_EQ
(
framework
::
product
(
left_tensor
->
dims
()),
framework
::
product
(
right_tensor
->
dims
()),
left_tensor
->
numel
(),
right_tensor
->
numel
(),
"Minus operator must take two tensor with same num of elements"
);
ctx
.
Output
<
framework
::
Tensor
>
(
"Out"
)
->
Resize
(
left_tensor
->
dims
());
}
...
...
paddle/operators/squared_l2_distance_op.cc
浏览文件 @
a2a69f2a
...
...
@@ -41,8 +41,7 @@ class SquaredL2DistanceOp : public framework::OperatorWithKernel {
int
rank
=
framework
::
arity
(
x_dims
);
PADDLE_ENFORCE_GE
(
rank
,
2
,
"Tensor rank should be at least equal to 2."
);
PADDLE_ENFORCE_EQ
(
framework
::
product
(
x_dims
)
/
x_dims
[
0
],
framework
::
product
(
y_dims
)
/
y_dims
[
0
],
PADDLE_ENFORCE_EQ
(
x
->
numel
()
/
x_dims
[
0
],
y
->
numel
()
/
y_dims
[
0
],
"Product of dimensions expcet the first dimension of "
"input and target must be equal."
);
PADDLE_ENFORCE
(
y_dims
[
0
]
==
1
||
y_dims
[
0
]
==
x_dims
[
0
],
...
...
@@ -50,8 +49,7 @@ class SquaredL2DistanceOp : public framework::OperatorWithKernel {
"or to 1."
);
ctx
.
Output
<
Tensor
>
(
"sub_result"
)
->
Resize
({
static_cast
<
int
>
(
x_dims
[
0
]),
static_cast
<
int
>
(
framework
::
product
(
x_dims
)
/
x_dims
[
0
])});
->
Resize
({
x_dims
[
0
],
x
->
numel
()
/
x_dims
[
0
]});
ctx
.
Output
<
Tensor
>
(
"Out"
)
->
Resize
({
x_dims
[
0
],
1
});
}
};
...
...
paddle/operators/squared_l2_distance_op.h
浏览文件 @
a2a69f2a
...
...
@@ -39,7 +39,7 @@ class SquaredL2DistanceKernel : public framework::OpKernel {
auto
in0_dims
=
in0
->
dims
();
auto
in1_dims
=
in1
->
dims
();
int
cols
=
framework
::
product
(
in0_dims
)
/
in0_dims
[
0
];
int
cols
=
in0
->
numel
(
)
/
in0_dims
[
0
];
// reduce dimensions except the first
auto
x
=
EigenMatrix
<
T
>::
From
(
*
in0
,
framework
::
make_ddim
({
in0_dims
[
0
],
cols
}));
...
...
@@ -82,7 +82,7 @@ class SquaredL2DistanceGradKernel : public framework::OpKernel {
auto
x_dims
=
x_g
->
dims
();
auto
y_dims
=
y_g
->
dims
();
int
cols
=
framework
::
product
(
x_dims
)
/
x_dims
[
0
];
int
cols
=
x_g
->
numel
(
)
/
x_dims
[
0
];
// calculate gradient
auto
grad_mat
=
2
*
(
out_grad
.
broadcast
(
Eigen
::
array
<
int
,
2
>
({{
1
,
cols
}})))
*
...
...
paddle/operators/uniform_random_op.cc
浏览文件 @
a2a69f2a
...
...
@@ -35,7 +35,7 @@ class CPUUniformRandomKernel : public framework::OpKernel {
std
::
uniform_real_distribution
<
T
>
dist
(
static_cast
<
T
>
(
context
.
Attr
<
float
>
(
"min"
)),
static_cast
<
T
>
(
context
.
Attr
<
float
>
(
"max"
)));
int64_t
size
=
framework
::
product
(
tensor
->
dims
()
);
int64_t
size
=
tensor
->
numel
(
);
for
(
int64_t
i
=
0
;
i
<
size
;
++
i
)
{
data
[
i
]
=
dist
(
engine
);
}
...
...
paddle/operators/uniform_random_op.cu
浏览文件 @
a2a69f2a
...
...
@@ -53,8 +53,8 @@ class GPUUniformRandomKernel : public framework::OpKernel {
T
min
=
static_cast
<
T
>
(
context
.
Attr
<
float
>
(
"min"
));
T
max
=
static_cast
<
T
>
(
context
.
Attr
<
float
>
(
"max"
));
thrust
::
counting_iterator
<
unsigned
int
>
index_sequence_begin
(
0
);
ssize_t
N
=
framework
::
product
(
tensor
->
dims
()
);
thrust
::
transform
(
index_sequence_begin
,
index_sequence_begin
+
N
,
int64_t
size
=
tensor
->
numel
(
);
thrust
::
transform
(
index_sequence_begin
,
index_sequence_begin
+
size
,
thrust
::
device_ptr
<
T
>
(
data
),
UniformGenerator
<
T
>
(
min
,
max
,
seed
));
}
...
...
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