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
7f00716c
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
7f00716c
编写于
3月 07, 2018
作者:
K
kexinzhao
提交者:
GitHub
3月 07, 2018
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Add context wait in type_transform (#8850)
上级
6f50dee4
变更
3
隐藏空白更改
内联
并排
Showing
3 changed file
with
33 addition
and
25 deletion
+33
-25
paddle/fluid/framework/data_type_transform.cc
paddle/fluid/framework/data_type_transform.cc
+1
-0
paddle/fluid/framework/data_type_transform_test.cc
paddle/fluid/framework/data_type_transform_test.cc
+12
-12
paddle/fluid/framework/data_type_transform_test.cu
paddle/fluid/framework/data_type_transform_test.cu
+20
-13
未找到文件。
paddle/fluid/framework/data_type_transform.cc
浏览文件 @
7f00716c
...
...
@@ -53,6 +53,7 @@ struct CastDataType {
auto
*
context
=
static_cast
<
const
platform
::
CUDADeviceContext
*>
(
ctx_
);
trans
(
*
context
,
in_begin
,
in_end
,
out_begin
,
CastDataTypeFunctor
<
InType
,
OutType
>
());
context
->
Wait
();
#endif
}
else
{
PADDLE_THROW
(
"Unsupported place!"
);
...
...
paddle/fluid/framework/data_type_transform_test.cc
浏览文件 @
7f00716c
...
...
@@ -50,13 +50,13 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType
(
kernel_fp32
,
kernel_fp64
,
in
,
&
out
);
double
*
out_data_double
=
out
.
data
<
double
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_double
[
i
],
static_cast
<
double
>
(
i
/
3
));
EXPEC
T_EQ
(
out_data_double
[
i
],
static_cast
<
double
>
(
i
/
3
));
}
TransDataType
(
kernel_fp32
,
kernel_int32
,
in
,
&
out
);
int
*
out_data_int
=
out
.
data
<
int
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_int
[
i
],
static_cast
<
int
>
(
i
/
3
));
EXPEC
T_EQ
(
out_data_int
[
i
],
static_cast
<
int
>
(
i
/
3
));
}
}
...
...
@@ -76,31 +76,31 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType
(
kernel_fp16
,
kernel_fp32
,
in
,
&
out
);
float
*
out_data_float
=
out
.
data
<
float
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_float
[
i
],
static_cast
<
float
>
(
ptr
[
i
]));
EXPEC
T_EQ
(
out_data_float
[
i
],
static_cast
<
float
>
(
ptr
[
i
]));
}
TransDataType
(
kernel_fp16
,
kernel_fp64
,
in
,
&
out
);
double
*
out_data_double
=
out
.
data
<
double
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_double
[
i
],
static_cast
<
double
>
(
ptr
[
i
]));
EXPEC
T_EQ
(
out_data_double
[
i
],
static_cast
<
double
>
(
ptr
[
i
]));
}
TransDataType
(
kernel_fp16
,
kernel_int32
,
in
,
&
out
);
int
*
out_data_int
=
out
.
data
<
int
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_int
[
i
],
static_cast
<
int
>
(
ptr
[
i
]));
EXPEC
T_EQ
(
out_data_int
[
i
],
static_cast
<
int
>
(
ptr
[
i
]));
}
TransDataType
(
kernel_fp16
,
kernel_int64
,
in
,
&
out
);
int64_t
*
out_data_int64
=
out
.
data
<
int64_t
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_int64
[
i
],
static_cast
<
int64_t
>
(
ptr
[
i
]));
EXPEC
T_EQ
(
out_data_int64
[
i
],
static_cast
<
int64_t
>
(
ptr
[
i
]));
}
TransDataType
(
kernel_fp16
,
kernel_bool
,
in
,
&
out
);
bool
*
out_data_bool
=
out
.
data
<
bool
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_bool
[
i
],
static_cast
<
bool
>
(
ptr
[
i
]));
EXPEC
T_EQ
(
out_data_bool
[
i
],
static_cast
<
bool
>
(
ptr
[
i
]));
}
// transform float to float16
...
...
@@ -112,7 +112,7 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType
(
kernel_fp32
,
kernel_fp16
,
in
,
&
out
);
ptr
=
out
.
data
<
float16
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_float
[
i
]).
x
);
EXPEC
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_float
[
i
]).
x
);
}
// transform double to float16
...
...
@@ -124,7 +124,7 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType
(
kernel_fp64
,
kernel_fp16
,
in
,
&
out
);
ptr
=
out
.
data
<
float16
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_double
[
i
]).
x
);
EXPEC
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_double
[
i
]).
x
);
}
// transform int to float16
...
...
@@ -136,7 +136,7 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType
(
kernel_int32
,
kernel_fp16
,
in
,
&
out
);
ptr
=
out
.
data
<
float16
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_int
[
i
]).
x
);
EXPEC
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_int
[
i
]).
x
);
}
// transform int64 to float16
...
...
@@ -148,7 +148,7 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType
(
kernel_int64
,
kernel_fp16
,
in
,
&
out
);
ptr
=
out
.
data
<
float16
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_int64
[
i
]).
x
);
EXPEC
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_int64
[
i
]).
x
);
}
// transform bool to float16
...
...
@@ -160,7 +160,7 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType
(
kernel_bool
,
kernel_fp16
,
in
,
&
out
);
ptr
=
out
.
data
<
float16
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_bool
[
i
]).
x
);
EXPEC
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_bool
[
i
]).
x
);
}
}
}
paddle/fluid/framework/data_type_transform_test.cu
浏览文件 @
7f00716c
...
...
@@ -49,15 +49,16 @@ TEST(DataTypeTransform, GPUTransform) {
float
arr
[
6
]
=
{
0
,
1
,
2
,
3
,
4
,
5
};
int
data_number
=
sizeof
(
arr
)
/
sizeof
(
arr
[
0
]);
memcpy
(
in_ptr
,
arr
,
sizeof
(
arr
));
TensorCopy
(
in
,
gpu_place
,
context
,
&
in_gpu
);
TensorCopy
(
in
,
gpu_place
,
context
,
&
in_gpu
);
context
.
Wait
();
TransDataType
(
kernel_fp32
,
kernel_fp64
,
in_gpu
,
&
out_gpu
);
TensorCopy
(
out_gpu
,
cpu_place
,
context
,
&
out
);
context
.
Wait
();
double
*
out_data_double
=
out
.
data
<
double
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_double
[
i
],
static_cast
<
double
>
(
arr
[
i
]));
EXPEC
T_EQ
(
out_data_double
[
i
],
static_cast
<
double
>
(
arr
[
i
]));
}
TransDataType
(
kernel_fp32
,
kernel_int32
,
in_gpu
,
&
out_gpu
);
...
...
@@ -66,7 +67,7 @@ TEST(DataTypeTransform, GPUTransform) {
int
*
out_data_int
=
out
.
data
<
int
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_int
[
i
],
static_cast
<
int
>
(
arr
[
i
]));
EXPEC
T_EQ
(
out_data_int
[
i
],
static_cast
<
int
>
(
arr
[
i
]));
}
}
...
...
@@ -83,6 +84,7 @@ TEST(DataTypeTransform, GPUTransform) {
int
data_number
=
sizeof
(
arr
)
/
sizeof
(
arr
[
0
]);
memcpy
(
ptr
,
arr
,
sizeof
(
arr
));
TensorCopy
(
in
,
gpu_place
,
context
,
&
in_gpu
);
context
.
Wait
();
// transform from float16 to other data types
TransDataType
(
kernel_fp16
,
kernel_fp32
,
in_gpu
,
&
out_gpu
);
...
...
@@ -91,7 +93,7 @@ TEST(DataTypeTransform, GPUTransform) {
float
*
out_data_float
=
out
.
data
<
float
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_float
[
i
],
static_cast
<
float
>
(
ptr
[
i
]));
EXPEC
T_EQ
(
out_data_float
[
i
],
static_cast
<
float
>
(
ptr
[
i
]));
}
TransDataType
(
kernel_fp16
,
kernel_fp64
,
in_gpu
,
&
out_gpu
);
...
...
@@ -100,7 +102,7 @@ TEST(DataTypeTransform, GPUTransform) {
double
*
out_data_double
=
out
.
data
<
double
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_double
[
i
],
static_cast
<
double
>
(
ptr
[
i
]));
EXPEC
T_EQ
(
out_data_double
[
i
],
static_cast
<
double
>
(
ptr
[
i
]));
}
TransDataType
(
kernel_fp16
,
kernel_int32
,
in_gpu
,
&
out_gpu
);
...
...
@@ -109,7 +111,7 @@ TEST(DataTypeTransform, GPUTransform) {
int
*
out_data_int
=
out
.
data
<
int
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_int
[
i
],
static_cast
<
int
>
(
ptr
[
i
]));
EXPEC
T_EQ
(
out_data_int
[
i
],
static_cast
<
int
>
(
ptr
[
i
]));
}
TransDataType
(
kernel_fp16
,
kernel_int64
,
in_gpu
,
&
out_gpu
);
...
...
@@ -118,7 +120,7 @@ TEST(DataTypeTransform, GPUTransform) {
int64_t
*
out_data_int64
=
out
.
data
<
int64_t
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_int64
[
i
],
static_cast
<
int64_t
>
(
ptr
[
i
]));
EXPEC
T_EQ
(
out_data_int64
[
i
],
static_cast
<
int64_t
>
(
ptr
[
i
]));
}
TransDataType
(
kernel_fp16
,
kernel_bool
,
in_gpu
,
&
out_gpu
);
...
...
@@ -127,7 +129,7 @@ TEST(DataTypeTransform, GPUTransform) {
bool
*
out_data_bool
=
out
.
data
<
bool
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
out_data_bool
[
i
],
static_cast
<
bool
>
(
ptr
[
i
]));
EXPEC
T_EQ
(
out_data_bool
[
i
],
static_cast
<
bool
>
(
ptr
[
i
]));
}
// transform float to float16
...
...
@@ -137,13 +139,14 @@ TEST(DataTypeTransform, GPUTransform) {
}
TensorCopy
(
in
,
gpu_place
,
context
,
&
in_gpu
);
context
.
Wait
();
TransDataType
(
kernel_fp32
,
kernel_fp16
,
in_gpu
,
&
out_gpu
);
TensorCopy
(
out_gpu
,
cpu_place
,
context
,
&
out
);
context
.
Wait
();
ptr
=
out
.
data
<
float16
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_float
[
i
]).
x
);
EXPEC
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_float
[
i
]).
x
);
}
// transform double to float16
...
...
@@ -154,13 +157,14 @@ TEST(DataTypeTransform, GPUTransform) {
}
TensorCopy
(
in
,
gpu_place
,
context
,
&
in_gpu
);
context
.
Wait
();
TransDataType
(
kernel_fp64
,
kernel_fp16
,
in_gpu
,
&
out_gpu
);
TensorCopy
(
out_gpu
,
cpu_place
,
context
,
&
out
);
context
.
Wait
();
ptr
=
out
.
data
<
float16
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_double
[
i
]).
x
);
EXPEC
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_double
[
i
]).
x
);
}
// transform int to float16
...
...
@@ -170,13 +174,14 @@ TEST(DataTypeTransform, GPUTransform) {
}
TensorCopy
(
in
,
gpu_place
,
context
,
&
in_gpu
);
context
.
Wait
();
TransDataType
(
kernel_int32
,
kernel_fp16
,
in_gpu
,
&
out_gpu
);
TensorCopy
(
out_gpu
,
cpu_place
,
context
,
&
out
);
context
.
Wait
();
ptr
=
out
.
data
<
float16
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_int
[
i
]).
x
);
EXPEC
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_int
[
i
]).
x
);
}
// transform int64 to float16
...
...
@@ -187,13 +192,14 @@ TEST(DataTypeTransform, GPUTransform) {
}
TensorCopy
(
in
,
gpu_place
,
context
,
&
in_gpu
);
context
.
Wait
();
TransDataType
(
kernel_int64
,
kernel_fp16
,
in_gpu
,
&
out_gpu
);
TensorCopy
(
out_gpu
,
cpu_place
,
context
,
&
out
);
context
.
Wait
();
ptr
=
out
.
data
<
float16
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_int64
[
i
]).
x
);
EXPEC
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_int64
[
i
]).
x
);
}
// transform bool to float16
...
...
@@ -203,13 +209,14 @@ TEST(DataTypeTransform, GPUTransform) {
}
TensorCopy
(
in
,
gpu_place
,
context
,
&
in_gpu
);
context
.
Wait
();
TransDataType
(
kernel_bool
,
kernel_fp16
,
in_gpu
,
&
out_gpu
);
TensorCopy
(
out_gpu
,
cpu_place
,
context
,
&
out
);
context
.
Wait
();
ptr
=
out
.
data
<
float16
>
();
for
(
int
i
=
0
;
i
<
data_number
;
++
i
)
{
ASSER
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_bool
[
i
]).
x
);
EXPEC
T_EQ
(
ptr
[
i
].
x
,
static_cast
<
float16
>
(
in_data_bool
[
i
]).
x
);
}
}
}
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录