Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
PaddlePaddle
Paddle
提交
eef55ca7
P
Paddle
项目概览
PaddlePaddle
/
Paddle
1 年多 前同步成功
通知
2302
Star
20931
Fork
5422
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1423
列表
看板
标记
里程碑
合并请求
543
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
Paddle
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
1,423
Issue
1,423
列表
看板
标记
里程碑
合并请求
543
合并请求
543
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
提交
eef55ca7
编写于
8月 03, 2017
作者:
Z
Zhuoyuan
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
remodify
上级
2b35fca1
变更
2
显示空白变更内容
内联
并排
Showing
2 changed file
with
108 addition
and
105 deletion
+108
-105
paddle/operators/gather_func.h
paddle/operators/gather_func.h
+41
-35
paddle/operators/scatter_func.h
paddle/operators/scatter_func.h
+67
-70
未找到文件。
paddle/operators/gather_func.h
浏览文件 @
eef55ca7
...
@@ -14,9 +14,9 @@ limitations under the License. */
...
@@ -14,9 +14,9 @@ limitations under the License. */
#pragma once
#pragma once
#include <cstring>
#include <cstring>
#include "paddle/framework/ddim.h"
#include "paddle/framework/tensor.h"
#include "paddle/framework/tensor.h"
#include "paddle/platform/place.h"
#include "paddle/platform/place.h"
#include "paddle/framework/ddim.h"
/**
/**
* Return a new tensor from source tensor, gathered according to index
* Return a new tensor from source tensor, gathered according to index
...
@@ -27,7 +27,7 @@ limitations under the License. */
...
@@ -27,7 +27,7 @@ limitations under the License. */
template
<
typename
Place
,
typename
T
>
template
<
typename
Place
,
typename
T
>
Tensor
*
Gather
(
Tensor
*
src
,
Tensor
*
index
)
{
Tensor
*
Gather
(
Tensor
*
src
,
Tensor
*
index
)
{
// check index of shape 1-D
// check index of shape 1-D
PADDLE_ENFORCE
(
index
->
dims
().
size
()
==
1
);
PADDLE_ENFORCE
(
index
->
dims
().
size
()
==
1
);
int
index_size
=
index
->
dims
()[
0
];
int
index_size
=
index
->
dims
()[
0
];
// Source shape
// Source shape
...
@@ -41,60 +41,66 @@ Tensor* Gather(Tensor* src, Tensor* index) {
...
@@ -41,60 +41,66 @@ Tensor* Gather(Tensor* src, Tensor* index) {
/* slice size */
/* slice size */
int
slice_size
=
1
;
int
slice_size
=
1
;
for
(
unsigned
int
i
=
0
;
i
<
src_dims
.
size
();
++
i
)
for
(
size_t
i
=
0
;
i
<
src_dims
.
size
();
++
i
)
slice_size
*=
src_dims
[
i
];
slice_size
*=
src_dims
[
i
];
/* Gathering */
/* Gathering */
if
(
place
==
CPUPlace
())
{
if
(
place
==
CPUPlace
())
{
// init for CPU
// init for CPU
output
=
New_tensor
.
mutable_data
<
T
>
(
output_dims
,
CPUPlace
());
output
=
New_tensor
.
mutable_data
<
T
>
(
output_dims
,
CPUPlace
());
CPUGather
(
src
->
data
(),
index
->
data
(),
slice_size
,
new_tensor
->
mutable_data
());
CPUGather
(
src
->
data
(),
index
->
data
(),
slice_size
,
new_tensor
->
mutable_data
());
}
else
{
// GPU
}
else
{
// GPU
// init for GPU
// init for GPU
output
=
New_tensor
.
mutable_data
<
T
>
(
output_dims
,
GPUPlace
());
output
=
New_tensor
.
mutable_data
<
T
>
(
output_dims
,
GPUPlace
());
/* how to specialize device??*/
/* how to specialize device??*/
GPUGather
(
d
,
src
->
data
(),
index
->
data
(),
slice_size
,
new_tensor
->
mutable_data
());
GPUGather
(
d
,
src
->
data
(),
index
->
data
(),
slice_size
,
new_tensor
->
mutable_data
());
}
}
return
New_tensor
;
return
New_tensor
;
}
}
/* Implementation of CPU copy */
/* Implementation of CPU copy */
template
<
typename
T
>
template
<
typename
T
>
void
CPUGather
(
const
T
*
params
,
const
int
*
indices
,
void
CPUGather
(
const
T
*
params
,
const
int
slice_size
,
const
int
index_size
,
const
int
*
indices
,
const
int
slice_size
,
const
int
index_size
,
T
*
output
)
{
T
*
output
)
{
const
size_t
slice_bytes
=
slice_size
*
sizeof
(
T
);
const
size_t
slice_bytes
=
slice_size
*
sizeof
(
T
);
for
(
int
i
=
0
;
i
<
index_size
;
++
i
)
for
(
size_t
i
=
0
;
i
<
index_size
;
++
i
)
{
int
index_
=
indices
[
i
];
int
index_
=
indices
[
i
];
/* copy src[index_] to output[i] */
/* copy src[index_] to output[i] */
memcpy
(
output
+
i
*
slice_bytes
,
memcpy
(
params
+
index_
*
slice_bytes
,
output
+
i
*
slice_bytes
,
params
+
index_
*
slice_bytes
,
slice_bytes
);
slice_bytes
);
}
}
}
/* Implementation of GPU copy:
/* Implementation of GPU copy:
I suppose the GPUDevice& d, contains gpu_id and thread_id
I suppose the GPUDevice& d, contains gpu_id and thread_id
d = cuda_stream(gpu_id_, stream_id_);
d = cuda_stream(gpu_id_, stream_id_);
*/
*/
template
<
typename
T
>
template
<
typename
T
>
void
GPUGather
(
const
GPUDevice
&
d
,
void
GPUGather
(
const
GPUDevice
&
d
,
const
T
*
src
,
const
int
*
index
,
const
T
*
src
,
const
int
slice_size
,
const
int
index_size
,
const
int
*
index
,
const
int
slice_size
,
const
int
index_size
,
T
*
output
)
{
T
*
output
)
{
int
block_count
=
slice_size
*
index_size
;
int
block_count
=
slice_size
*
index_size
;
int
thread_per_block
=
1024
;
int
thread_per_block
=
1024
;
GatherOpKernel
<
T
>
GatherOpKernel
<
T
><<<
block_count
,
thread_per_block
,
0
,
d
.
stream
()
>>>
(
<<<
block_count
,
thread_per_block
,
0
,
d
.
stream
()
>>>
(
src
,
index
,
output
,
slice_size
,
indices_size
,
slice_size
,
out_size
);
src
,
index
,
output
,
slice_size
,
indices_size
,
slice_size
,
out_size
);
}
}
template
<
typename
T
>
template
<
typename
T
>
__global__
void
GatherOpKernel
(
const
T
*
params
,
const
int
*
indices
,
T
*
out
,
__global__
void
GatherOpKernel
(
const
T
*
params
,
const
int
*
indices
,
T
*
out
,
int64
indices_size
,
int64
indices_size
,
int64
slice_size
,
int64
out_size
)
{
int64
slice_size
,
int64
out_size
)
{
/* I suppose we have the following macro,
/* I suppose we have the following macro,
which I strongly suggest that we should put in cuda:
which I strongly suggest that we should put in cuda:
#define CUDA_1D_KERNEL_LOOP(i, n) \
#define CUDA_1D_KERNEL_LOOP(i, n) \
...
...
paddle/operators/scatter_func.h
浏览文件 @
eef55ca7
...
@@ -14,95 +14,92 @@ limitations under the License. */
...
@@ -14,95 +14,92 @@ limitations under the License. */
#pragma once
#pragma once
#include <cstring>
#include <cstring>
#include "paddle/framework/ddim.h"
#include "paddle/framework/tensor.h"
#include "paddle/framework/tensor.h"
#include "paddle/platform/place.h"
#include "paddle/platform/place.h"
#include "paddle/framework/ddim.h"
/**
/**
* Return a updated tensor from source tensor, scattered according to index:
* Return a updated tensor from source tensor, scattered according to index:
* dst[i] += src[index[i]]
* dst[i] += src[index[i]]
* input[src]: type-T source Tensor
* input[src]: type-T source Tensor
* input[
I
ndex]: type-int index Tensor (1-D)
* input[
i
ndex]: type-int index Tensor (1-D)
* return: output tensor
* return: output tensor
*/
*/
template
<
typename
place
,
typename
T
>
template
<
typename
Place
,
typename
T
>
void
ScatterUpdate_func
(
Tensor
*
Src
,
Tensor
*
Dst
,
Tensor
*
Index
)
{
void
ScatterUpdate
(
Tensor
*
src
,
Tensor
*
dst
,
Tensor
*
index
)
{
// assert index is an int-type tensor
assert
(
Index
->
istype
(
int
));
// Source shape
// Source shape
auto
src_dims
=
S
rc
->
dims
();
auto
src_dims
=
s
rc
->
dims
();
auto
dst_dims
=
D
st
->
dims
();
auto
dst_dims
=
d
st
->
dims
();
DDim
output_dims
(
dims_src
);
DDim
output_dims
(
dims_src
);
// check Src shape and D
st shape should match
// check src shape and d
st shape should match
for
(
in
t
i
=
1
;
i
<
src_dims
.
size
();
i
++
)
for
(
size_
t
i
=
1
;
i
<
src_dims
.
size
();
i
++
)
assert
(
src_dims
[
i
]
==
dst_dims
[
i
]);
PADDLE_ENFORCE
(
src_dims
[
i
]
==
dst_dims
[
i
]);
int
index_size
=
I
ndex
->
dims
()[
0
];
int
index_size
=
i
ndex
->
dims
()[
0
];
/* slice size */
/* slice size */
int
slice_size
=
1
;
int
slice_size
=
1
;
for
(
unsigned
int
i
=
0
;
i
<
src_dims
.
size
();
++
i
)
for
(
size_t
i
=
0
;
i
<
src_dims
.
size
();
++
i
)
slice_size
*=
src_dims
[
i
];
slice_size
*=
src_dims
[
i
];
if
(
place
==
CPUPlace
())
{
if
(
place
==
CPUPlace
())
{
// init
// init
output
=
new_tensor
.
mutable_data
<
T
>
(
output_dims
,
CPUPlace
());
output
=
new_tensor
.
mutable_data
<
T
>
(
output_dims
,
CPUPlace
());
CPUScatterUpdate
(
src
->
data
(),
index
->
data
(),
slice_size
,
new_tensor
->
mutable_data
());
CPUScatterUpdate
(
src
->
data
(),
index
->
data
(),
slice_size
,
new_tensor
->
mutable_data
());
}
else
{
// GPU
}
else
{
// GPU
// init
// init
output
=
new_tensor
.
mutable_data
<
T
>
(
output_dims
,
GPUPlace
());
output
=
new_tensor
.
mutable_data
<
T
>
(
output_dims
,
GPUPlace
());
/* how to specialize device??*/
/* how to specialize device??*/
GPUScatterUpdate
(
d
,
src
->
data
(),
index
->
data
(),
slice_size
,
new_tensor
->
mutable_data
());
GPUScatterUpdate
(
d
,
src
->
data
(),
index
->
data
(),
slice_size
,
new_tensor
->
mutable_data
());
}
}
}
}
/* Implementation of CPU copy */
/* Implementation of CPU copy */
template
<
typename
T
>
template
<
typename
T
>
void
CPUScatterUpdate
(
const
T
*
src
,
const
int
*
Index
,
void
CPUScatterUpdate
(
const
T
*
src
,
const
int
slice_size
,
const
int
index_size
,
const
int
*
index
,
const
int
slice_size
,
const
int
index_size
,
T
*
output
)
{
T
*
output
)
{
//const size_t slice_bytes = slice_size * sizeof(T);
//
const size_t slice_bytes = slice_size * sizeof(T);
for
(
int
i
=
0
;
i
<
index_size
;
++
i
)
for
(
size_t
i
=
0
;
i
<
index_size
;
++
i
)
{
int
index_
=
index
[
i
];
int
index_
=
index
[
i
];
/* dst[index_] += src[index_]
math
::
vAdd
<
T
>
(
slice_size
,
add operation size: slice_size
src
+
index_
*
slice_bytes
,
*/
math
::
vAdd
<
T
>
(
slice_size
,
src
+
index_
*
slice_bytes
,
output
+
i
*
slice_bytes
,
output
+
i
*
slice_bytes
,
output
+
i
*
slice_bytes
);
output
+
i
*
slice_bytes
);
/* Scatter update, not just assign
}
memcpy(output + i * slice_bytes,
src + index_ * slice_bytes,
slice_bytes);
*/
}
}
/* Implementation of GPU scatter:
/* Implementation of GPU scatter:
I suppose the GPUDevice& d, contains gpu_id and thread_id
I suppose the GPUDevice& d, contains gpu_id and thread_id
d = cuda_stream(gpu_id_, stream_id_);
d = cuda_stream(gpu_id_, stream_id_);
*/
*/
template
<
typename
T
>
template
<
typename
T
>
void
GPUScatterUpdate
(
const
GPUDevice
&
d
,
void
GPUScatterUpdate
(
const
GPUDevice
&
d
,
const
T
*
src
,
const
int
*
Index
,
const
T
*
src
,
const
int
slice_size
,
const
int
index_size
,
const
int
*
index
,
const
int
slice_size
,
const
int
index_size
,
T
*
output
)
{
T
*
output
)
{
int
block_count
=
slice_size
*
index_size
;
int
block_count
=
slice_size
*
index_size
;
int
thread_per_block
=
1024
;
int
thread_per_block
=
1024
;
ScatterOpKernel
<
T
>
ScatterOpKernel
<
T
><<<
block_count
,
thread_per_block
,
0
,
d
.
stream
()
>>>
(
<<<
block_count
,
thread_per_block
,
0
,
d
.
stream
()
>>>
(
src
,
index
,
output
,
slice_size
,
indices_size
,
slice_size
,
out_size
);
src
,
Index
,
output
,
slice_size
,
indices_size
,
slice_size
,
out_size
);
}
}
template
<
typename
T
>
template
<
typename
T
>
__global__
void
ScatterOpKernel
(
const
T
*
params
,
const
int
*
indices
,
T
*
out
,
__global__
void
ScatterOpKernel
(
const
T
*
params
,
const
int
*
indices
,
T
*
out
,
int64
indices_size
,
int64
indices_size
,
int64
slice_size
,
int64
out_size
)
{
int64
slice_size
,
int64
out_size
)
{
/* I suppose we have the following macro,
/* I suppose we have the following macro,
which I strongly suggest that we should put in cuda:
which I strongly suggest that we should put in cuda:
#define CUDA_1D_KERNEL_LOOP(i, n) \
#define CUDA_1D_KERNEL_LOOP(i, n) \
...
...
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