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
PaddleDetection
提交
c1e9b1e3
P
PaddleDetection
项目概览
PaddlePaddle
/
PaddleDetection
大约 1 年 前同步成功
通知
695
Star
11112
Fork
2696
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
184
列表
看板
标记
里程碑
合并请求
40
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
PaddleDetection
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
184
Issue
184
列表
看板
标记
里程碑
合并请求
40
合并请求
40
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
未验证
提交
c1e9b1e3
编写于
3月 20, 2018
作者:
K
Kexin Zhao
提交者:
GitHub
3月 20, 2018
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #9231 from kexinzhao/elementwise_add_fp16
Add float16 support to Elementwise Add op
上级
d126933d
d307b5e4
变更
3
显示空白变更内容
内联
并排
Showing
3 changed file
with
230 addition
and
146 deletion
+230
-146
paddle/fluid/operators/elementwise_add_op.cu
paddle/fluid/operators/elementwise_add_op.cu
+11
-10
paddle/fluid/platform/float16.h
paddle/fluid/platform/float16.h
+30
-32
python/paddle/fluid/tests/unittests/test_elementwise_add_op.py
...n/paddle/fluid/tests/unittests/test_elementwise_add_op.py
+189
-104
未找到文件。
paddle/fluid/operators/elementwise_add_op.cu
浏览文件 @
c1e9b1e3
...
...
@@ -14,19 +14,20 @@ limitations under the License. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/elementwise_add_op.h"
#include "paddle/fluid/platform/float16.h"
namespace
ops
=
paddle
::
operators
;
namespace
plat
=
paddle
::
platform
;
REGISTER_OP_CUDA_KERNEL
(
elementwise_add
,
ops
::
ElementwiseAddKernel
<
p
addle
::
platform
::
CUDADeviceContext
,
float
>
,
ops
::
ElementwiseAddKernel
<
p
addle
::
platform
::
CUDADeviceContext
,
double
>
,
ops
::
ElementwiseAddKernel
<
p
addle
::
platform
::
CUDADeviceContext
,
in
t
>
,
ops
::
ElementwiseAddKernel
<
p
addle
::
platform
::
CUDADeviceContext
,
int64_t
>
);
elementwise_add
,
ops
::
ElementwiseAddKernel
<
plat
::
CUDADeviceContext
,
float
>
,
ops
::
ElementwiseAddKernel
<
p
lat
::
CUDADeviceContext
,
double
>
,
ops
::
ElementwiseAddKernel
<
p
lat
::
CUDADeviceContext
,
int
>
,
ops
::
ElementwiseAddKernel
<
p
lat
::
CUDADeviceContext
,
int64_
t
>
,
ops
::
ElementwiseAddKernel
<
p
lat
::
CUDADeviceContext
,
plat
::
float16
>
);
REGISTER_OP_CUDA_KERNEL
(
elementwise_add_grad
,
ops
::
ElementwiseAddGradKernel
<
paddle
::
platform
::
CUDADeviceContext
,
float
>
,
ops
::
ElementwiseAddGradKernel
<
paddle
::
platform
::
CUDADeviceContext
,
double
>
,
ops
::
ElementwiseAddGradKernel
<
paddle
::
platform
::
CUDADeviceContext
,
int
>
,
ops
::
ElementwiseAddGradKernel
<
paddle
::
platform
::
CUDADeviceContext
,
int64_t
>
);
ops
::
ElementwiseAddGradKernel
<
plat
::
CUDADeviceContext
,
float
>
,
ops
::
ElementwiseAddGradKernel
<
plat
::
CUDADeviceContext
,
double
>
,
ops
::
ElementwiseAddGradKernel
<
plat
::
CUDADeviceContext
,
int
>
,
ops
::
ElementwiseAddGradKernel
<
plat
::
CUDADeviceContext
,
int64_t
>
);
paddle/fluid/platform/float16.h
浏览文件 @
c1e9b1e3
...
...
@@ -600,7 +600,7 @@ HOSTDEVICE inline bool operator>=(const float16& a, const float16& b) {
// Arithmetic operators for float16 on ARMv8.2-A CPU
#elif defined(PADDLE_WITH_NATIVE_FP16)
HOST
inline
float16
operator
+
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
float16
operator
+
(
const
float16
&
a
,
const
float16
&
b
)
{
float16
res
;
asm
volatile
(
"ld1 {v0.h}[0], [%[a_ptr]]
\n
"
...
...
@@ -616,7 +616,7 @@ HOST inline float16 operator+(const float16& a, const float16& b) {
return
res
;
}
HOST
inline
float16
operator
-
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
float16
operator
-
(
const
float16
&
a
,
const
float16
&
b
)
{
float16
res
;
asm
volatile
(
"ld1 {v0.h}[0], [%[a_ptr]]
\n
"
...
...
@@ -632,7 +632,7 @@ HOST inline float16 operator-(const float16& a, const float16& b) {
return
res
;
}
HOST
inline
float16
operator
*
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
float16
operator
*
(
const
float16
&
a
,
const
float16
&
b
)
{
float16
res
;
asm
volatile
(
"ld1 {v0.h}[0], [%[a_ptr]]
\n
"
...
...
@@ -648,7 +648,7 @@ HOST inline float16 operator*(const float16& a, const float16& b) {
return
res
;
}
HOST
inline
float16
operator
/
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
float16
operator
/
(
const
float16
&
a
,
const
float16
&
b
)
{
float16
res
;
asm
volatile
(
"ld1 {v0.h}[0], [%[a_ptr]]
\n
"
...
...
@@ -664,7 +664,7 @@ HOST inline float16 operator/(const float16& a, const float16& b) {
return
res
;
}
HOST
inline
float16
operator
-
(
const
float16
&
a
)
{
inline
float16
operator
-
(
const
float16
&
a
)
{
float16
res
;
asm
volatile
(
"ld1 {v0.h}[0], [%[a_ptr]]
\n
"
...
...
@@ -679,27 +679,27 @@ HOST inline float16 operator-(const float16& a) {
return
res
;
}
HOST
inline
float16
&
operator
+=
(
float16
&
a
,
const
float16
&
b
)
{
inline
float16
&
operator
+=
(
float16
&
a
,
const
float16
&
b
)
{
a
=
a
+
b
;
return
a
;
}
HOST
inline
float16
&
operator
-=
(
float16
&
a
,
const
float16
&
b
)
{
inline
float16
&
operator
-=
(
float16
&
a
,
const
float16
&
b
)
{
a
=
a
-
b
;
return
a
;
}
HOST
inline
float16
&
operator
*=
(
float16
&
a
,
const
float16
&
b
)
{
inline
float16
&
operator
*=
(
float16
&
a
,
const
float16
&
b
)
{
a
=
a
*
b
;
return
a
;
}
HOST
inline
float16
&
operator
/=
(
float16
&
a
,
const
float16
&
b
)
{
inline
float16
&
operator
/=
(
float16
&
a
,
const
float16
&
b
)
{
a
=
a
/
b
;
return
a
;
}
HOST
inline
bool
operator
==
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
bool
operator
==
(
const
float16
&
a
,
const
float16
&
b
)
{
uint16_t
res
;
asm
volatile
(
"ld1 {v0.h}[0], [%[a_ptr]]
\n
"
...
...
@@ -715,11 +715,9 @@ HOST inline bool operator==(const float16& a, const float16& b) {
return
(
res
&
0xffff
)
!=
0
;
}
HOST
inline
bool
operator
!=
(
const
float16
&
a
,
const
float16
&
b
)
{
return
!
(
a
==
b
);
}
inline
bool
operator
!=
(
const
float16
&
a
,
const
float16
&
b
)
{
return
!
(
a
==
b
);
}
HOST
inline
bool
operator
<
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
bool
operator
<
(
const
float16
&
a
,
const
float16
&
b
)
{
uint16_t
res
;
asm
volatile
(
"ld1 {v1.h}[0], [%[a_ptr]]
\n
"
...
...
@@ -735,7 +733,7 @@ HOST inline bool operator<(const float16& a, const float16& b) {
return
(
res
&
0xffff
)
!=
0
;
}
HOST
inline
bool
operator
<=
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
bool
operator
<=
(
const
float16
&
a
,
const
float16
&
b
)
{
uint16_t
res
;
asm
volatile
(
"ld1 {v1.h}[0], [%[a_ptr]]
\n
"
...
...
@@ -751,7 +749,7 @@ HOST inline bool operator<=(const float16& a, const float16& b) {
return
(
res
&
0xffff
)
!=
0
;
}
HOST
inline
bool
operator
>
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
bool
operator
>
(
const
float16
&
a
,
const
float16
&
b
)
{
uint16_t
res
;
asm
volatile
(
"ld1 {v0.h}[0], [%[a_ptr]]
\n
"
...
...
@@ -767,7 +765,7 @@ HOST inline bool operator>(const float16& a, const float16& b) {
return
(
res
&
0xffff
)
!=
0
;
}
HOST
inline
bool
operator
>=
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
bool
operator
>=
(
const
float16
&
a
,
const
float16
&
b
)
{
uint16_t
res
;
asm
volatile
(
"ld1 {v0.h}[0], [%[a_ptr]]
\n
"
...
...
@@ -785,69 +783,69 @@ HOST inline bool operator>=(const float16& a, const float16& b) {
// Arithmetic operators for float16, software emulated on other CPU
#else
HOST
inline
float16
operator
+
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
float16
operator
+
(
const
float16
&
a
,
const
float16
&
b
)
{
return
float16
(
float
(
a
)
+
float
(
b
));
}
HOST
inline
float16
operator
-
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
float16
operator
-
(
const
float16
&
a
,
const
float16
&
b
)
{
return
float16
(
float
(
a
)
-
float
(
b
));
}
HOST
inline
float16
operator
*
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
float16
operator
*
(
const
float16
&
a
,
const
float16
&
b
)
{
return
float16
(
float
(
a
)
*
float
(
b
));
}
HOST
inline
float16
operator
/
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
float16
operator
/
(
const
float16
&
a
,
const
float16
&
b
)
{
return
float16
(
float
(
a
)
/
float
(
b
));
}
HOST
inline
float16
operator
-
(
const
float16
&
a
)
{
inline
float16
operator
-
(
const
float16
&
a
)
{
float16
res
;
res
.
x
=
a
.
x
^
0x8000
;
return
res
;
}
HOST
inline
float16
&
operator
+=
(
float16
&
a
,
const
float16
&
b
)
{
inline
float16
&
operator
+=
(
float16
&
a
,
const
float16
&
b
)
{
a
=
float16
(
float
(
a
)
+
float
(
b
));
return
a
;
}
HOST
inline
float16
&
operator
-=
(
float16
&
a
,
const
float16
&
b
)
{
inline
float16
&
operator
-=
(
float16
&
a
,
const
float16
&
b
)
{
a
=
float16
(
float
(
a
)
-
float
(
b
));
return
a
;
}
HOST
inline
float16
&
operator
*=
(
float16
&
a
,
const
float16
&
b
)
{
inline
float16
&
operator
*=
(
float16
&
a
,
const
float16
&
b
)
{
a
=
float16
(
float
(
a
)
*
float
(
b
));
return
a
;
}
HOST
inline
float16
&
operator
/=
(
float16
&
a
,
const
float16
&
b
)
{
inline
float16
&
operator
/=
(
float16
&
a
,
const
float16
&
b
)
{
a
=
float16
(
float
(
a
)
/
float
(
b
));
return
a
;
}
HOST
inline
bool
operator
==
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
bool
operator
==
(
const
float16
&
a
,
const
float16
&
b
)
{
return
float
(
a
)
==
float
(
b
);
}
HOST
inline
bool
operator
!=
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
bool
operator
!=
(
const
float16
&
a
,
const
float16
&
b
)
{
return
float
(
a
)
!=
float
(
b
);
}
HOST
inline
bool
operator
<
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
bool
operator
<
(
const
float16
&
a
,
const
float16
&
b
)
{
return
float
(
a
)
<
float
(
b
);
}
HOST
inline
bool
operator
<=
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
bool
operator
<=
(
const
float16
&
a
,
const
float16
&
b
)
{
return
float
(
a
)
<=
float
(
b
);
}
HOST
inline
bool
operator
>
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
bool
operator
>
(
const
float16
&
a
,
const
float16
&
b
)
{
return
float
(
a
)
>
float
(
b
);
}
HOST
inline
bool
operator
>=
(
const
float16
&
a
,
const
float16
&
b
)
{
inline
bool
operator
>=
(
const
float16
&
a
,
const
float16
&
b
)
{
return
float
(
a
)
>=
float
(
b
);
}
#endif
...
...
python/paddle/fluid/tests/unittests/test_elementwise_add_op.py
浏览文件 @
c1e9b1e3
...
...
@@ -13,158 +13,243 @@
# limitations under the License.
import
unittest
import
numpy
as
np
import
paddle.fluid.core
as
core
from
op_test
import
OpTest
class
TestElementwiseOp
(
OpTest
):
class
TestElementwise
Add
Op
(
OpTest
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_add"
self
.
dtype
=
np
.
float32
self
.
axis
=
-
1
self
.
init_dtype
()
self
.
init_input_output
()
self
.
init_axis
()
self
.
inputs
=
{
'X'
:
np
.
random
.
uniform
(
0.1
,
1
,
[
13
,
17
]).
astype
(
"float32"
),
'Y'
:
np
.
random
.
uniform
(
0.1
,
1
,
[
13
,
17
]).
astype
(
"float32"
)
'X'
:
OpTest
.
np_dtype_to_fluid_dtype
(
self
.
x
),
'Y'
:
OpTest
.
np_dtype_to_fluid_dtype
(
self
.
y
)
}
self
.
outputs
=
{
'Out'
:
np
.
add
(
self
.
inputs
[
'X'
],
self
.
inputs
[
'Y'
])}
self
.
attrs
=
{
'axis'
:
self
.
axis
}
self
.
outputs
=
{
'Out'
:
self
.
out
}
def
test_check_output
(
self
):
self
.
check_output
()
def
test_check_grad_normal
(
self
):
if
self
.
dtype
==
np
.
float16
:
return
self
.
check_grad
([
'X'
,
'Y'
],
'Out'
,
max_relative_error
=
0.005
)
def
test_check_grad_ingore_x
(
self
):
if
self
.
dtype
==
np
.
float16
:
return
self
.
check_grad
(
[
'Y'
],
'Out'
,
max_relative_error
=
0.005
,
no_grad_set
=
set
(
"X"
))
def
test_check_grad_ingore_y
(
self
):
if
self
.
dtype
==
np
.
float16
:
return
self
.
check_grad
(
[
'X'
],
'Out'
,
max_relative_error
=
0.005
,
no_grad_set
=
set
(
'Y'
))
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
uniform
(
0.1
,
1
,
[
13
,
17
]).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
uniform
(
0.1
,
1
,
[
13
,
17
]).
astype
(
self
.
dtype
)
self
.
out
=
np
.
add
(
self
.
x
,
self
.
y
)
class
TestElementwiseAddOp_scalar
(
TestElementwiseOp
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_add"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float32
),
'Y'
:
np
.
random
.
rand
(
1
).
astype
(
np
.
float32
)
}
self
.
outputs
=
{
'Out'
:
self
.
inputs
[
'X'
]
+
self
.
inputs
[
'Y'
]}
def
init_dtype
(
self
):
pass
def
init_axis
(
self
):
pass
class
TestElementwiseAddOp_scalar2
(
TestElementwiseOp
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_add"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float32
),
'Y'
:
np
.
random
.
rand
(
1
,
1
).
astype
(
np
.
float32
)
}
self
.
outputs
=
{
'Out'
:
self
.
inputs
[
'X'
]
+
self
.
inputs
[
'Y'
]}
class
TestFP16ElementwiseAddOp
(
TestElementwiseAddOp
):
def
init_dtype
(
self
):
self
.
dtype
=
np
.
float16
class
TestElementwiseAddOp_Vector
(
TestElementwiseOp
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_add"
self
.
inputs
=
{
'X'
:
np
.
random
.
random
((
32
,
)).
astype
(
"float32"
),
'Y'
:
np
.
random
.
random
((
32
,
)).
astype
(
"float32"
)
}
self
.
outputs
=
{
'Out'
:
np
.
add
(
self
.
inputs
[
'X'
],
self
.
inputs
[
'Y'
])}
def
test_check_output
(
self
):
if
core
.
is_compiled_with_cuda
():
place
=
core
.
CUDAPlace
(
0
)
if
core
.
is_float16_supported
(
place
):
self
.
check_output_with_place
(
place
,
atol
=
1e-3
)
class
TestElementwiseAddOp_broadcast_0
(
TestElementwiseOp
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_add"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float32
),
'Y'
:
np
.
random
.
rand
(
2
).
astype
(
np
.
float32
)
}
class
TestElementwiseAddOp_scalar
(
TestElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
1
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
self
.
attrs
=
{
'axis'
:
0
}
self
.
outputs
=
{
'Out'
:
self
.
inputs
[
'X'
]
+
self
.
inputs
[
'Y'
].
reshape
(
2
,
1
,
1
)
}
class
TestFP16ElementwiseAddOp_scalar
(
TestFP16ElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
1
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
class
TestElementwiseAddOp_broadcast_1
(
TestElementwiseOp
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_add"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float32
),
'Y'
:
np
.
random
.
rand
(
3
).
astype
(
np
.
float32
)
}
self
.
attrs
=
{
'axis'
:
1
}
self
.
outputs
=
{
'Out'
:
self
.
inputs
[
'X'
]
+
self
.
inputs
[
'Y'
].
reshape
(
1
,
3
,
1
)
}
class
TestElementwiseAddOp_scalar2
(
TestElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
1
,
1
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
class
TestElementwiseAddOp_broadcast_2
(
TestElementwiseOp
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_add"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float32
),
'Y'
:
np
.
random
.
rand
(
4
).
astype
(
np
.
float32
)
}
class
TestFP16ElementwiseAddOp_scalar2
(
TestFP16ElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
1
,
1
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
self
.
outputs
=
{
'Out'
:
self
.
inputs
[
'X'
]
+
self
.
inputs
[
'Y'
].
reshape
(
1
,
1
,
4
)
}
class
TestElementwiseAddOp_Vector
(
TestElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
random
((
32
,
)).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
random
((
32
,
)).
astype
(
self
.
dtype
)
self
.
out
=
np
.
add
(
self
.
x
,
self
.
y
)
class
TestElementwiseAddOp_broadcast_3
(
TestElementwiseOp
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_add"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
,
5
).
astype
(
np
.
float32
),
'Y'
:
np
.
random
.
rand
(
3
,
4
).
astype
(
np
.
float32
)
}
self
.
attrs
=
{
'axis'
:
1
}
self
.
outputs
=
{
'Out'
:
self
.
inputs
[
'X'
]
+
self
.
inputs
[
'Y'
].
reshape
(
1
,
3
,
4
,
1
)
}
class
TestFP16ElementwiseAddOp_Vector
(
TestFP16ElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
random
((
32
,
)).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
random
((
32
,
)).
astype
(
self
.
dtype
)
self
.
out
=
np
.
add
(
self
.
x
,
self
.
y
)
class
TestElementwiseAddOp_broadcast_4
(
TestElementwiseOp
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_add"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
,
5
).
astype
(
np
.
float32
),
'Y'
:
np
.
random
.
rand
(
2
,
1
).
astype
(
np
.
float32
)
}
class
TestElementwiseAddOp_broadcast_0
(
TestElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
2
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
2
,
1
,
1
)
self
.
attrs
=
{
'axis'
:
0
}
self
.
outputs
=
{
'Out'
:
self
.
inputs
[
'X'
]
+
self
.
inputs
[
'Y'
].
reshape
(
2
,
1
,
1
,
1
)
}
def
init_axis
(
self
):
self
.
axis
=
0
class
TestElementwiseAddOp_rowwise_add_0
(
TestElementwiseOp
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_add"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float32
),
'Y'
:
np
.
random
.
rand
(
3
,
4
).
astype
(
np
.
float32
)
}
class
TestFP16ElementwiseAddOp_broadcast_0
(
TestFP16ElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
2
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
2
,
1
,
1
)
self
.
attrs
=
{
'axis'
:
1
}
self
.
outputs
=
{
'Out'
:
self
.
inputs
[
'X'
]
+
self
.
inputs
[
'Y'
].
reshape
(
1
,
3
,
4
)
}
def
init_axis
(
self
):
self
.
axis
=
0
class
TestElementwiseAddOp_rowwise_add_1
(
TestElementwiseOp
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_add"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
1
).
astype
(
np
.
float32
),
'Y'
:
np
.
random
.
rand
(
1
).
astype
(
np
.
float32
)
}
class
TestElementwiseAddOp_broadcast_1
(
TestElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
3
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
1
,
3
,
1
)
self
.
attrs
=
{
'axis'
:
1
}
self
.
outputs
=
{
'Out'
:
self
.
inputs
[
'X'
]
+
self
.
inputs
[
'Y'
].
reshape
(
1
,
1
)
}
def
init_axis
(
self
):
self
.
axis
=
1
class
TestFP16ElementwiseAddOp_broadcast_1
(
TestFP16ElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
3
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
1
,
3
,
1
)
def
init_axis
(
self
):
self
.
axis
=
1
class
TestElementwiseAddOp_broadcast_2
(
TestElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
4
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
1
,
1
,
4
)
class
TestFP16ElementwiseAddOp_broadcast_2
(
TestFP16ElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
4
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
1
,
1
,
4
)
class
TestElementwiseAddOp_broadcast_3
(
TestElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
,
5
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
3
,
4
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
1
,
3
,
4
,
1
)
def
init_axis
(
self
):
self
.
axis
=
1
class
TestFP16ElementwiseAddOp_broadcast_3
(
TestFP16ElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
,
5
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
3
,
4
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
1
,
3
,
4
,
1
)
def
init_axis
(
self
):
self
.
axis
=
1
class
TestElementwiseAddOp_broadcast_4
(
TestElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
,
5
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
2
,
1
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
2
,
1
,
1
,
1
)
def
init_axis
(
self
):
self
.
axis
=
0
class
TestFP16ElementwiseAddOp_broadcast_4
(
TestFP16ElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
,
5
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
2
,
1
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
2
,
1
,
1
,
1
)
def
init_axis
(
self
):
self
.
axis
=
0
class
TestElementwiseAddOp_rowwise_add_0
(
TestElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
3
,
4
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
1
,
3
,
4
)
def
init_axis
(
self
):
self
.
axis
=
1
class
TestFP16ElementwiseAddOp_rowwise_add_0
(
TestFP16ElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
3
,
4
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
1
,
3
,
4
)
def
init_axis
(
self
):
self
.
axis
=
1
class
TestElementwiseAddOp_rowwise_add_1
(
TestElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
1
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
1
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
1
,
1
)
def
init_axis
(
self
):
self
.
axis
=
1
class
TestFP16ElementwiseAddOp_rowwise_add_1
(
TestFP16ElementwiseAddOp
):
def
init_input_output
(
self
):
self
.
x
=
np
.
random
.
rand
(
2
,
1
).
astype
(
self
.
dtype
)
self
.
y
=
np
.
random
.
rand
(
1
).
astype
(
self
.
dtype
)
self
.
out
=
self
.
x
+
self
.
y
.
reshape
(
1
,
1
)
def
init_axis
(
self
):
self
.
axis
=
1
if
__name__
==
'__main__'
:
...
...
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