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
xuexixianjinzhishi
Paddle
提交
745477fe
P
Paddle
项目概览
xuexixianjinzhishi
/
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 >>
未验证
提交
745477fe
编写于
12月 23, 2021
作者:
Y
yaoxuefeng
提交者:
GitHub
12月 23, 2021
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
add mem pool (#37127)
add mem pool
上级
6b59b58c
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
114 addition
and
1 deletion
+114
-1
paddle/fluid/framework/fleet/heter_ps/CMakeLists.txt
paddle/fluid/framework/fleet/heter_ps/CMakeLists.txt
+1
-1
paddle/fluid/framework/fleet/heter_ps/mem_pool.h
paddle/fluid/framework/fleet/heter_ps/mem_pool.h
+113
-0
未找到文件。
paddle/fluid/framework/fleet/heter_ps/CMakeLists.txt
浏览文件 @
745477fe
...
...
@@ -7,7 +7,7 @@ IF(WITH_GPU)
get_property
(
RPC_DEPS GLOBAL PROPERTY RPC_DEPS
)
SET
(
HETERPS_DEPS
${
HETERPS_DEPS
}
${
RPC_DEPS
}
)
endif
()
nv_library
(
heter_comm SRCS heter_comm.h feature_value.h heter_resource.cc heter_resource.h hashtable.h DEPS
${
HETERPS_DEPS
}
)
nv_library
(
heter_comm SRCS heter_comm.h feature_value.h heter_resource.cc heter_resource.h hashtable.h
mem_pool.h
DEPS
${
HETERPS_DEPS
}
)
nv_test
(
test_heter_comm SRCS feature_value.h DEPS heter_comm
)
nv_library
(
heter_ps SRCS heter_ps.cu DEPS heter_comm
)
ENDIF
()
...
...
paddle/fluid/framework/fleet/heter_ps/mem_pool.h
0 → 100644
浏览文件 @
745477fe
/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
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_WITH_HETERPS
// #include
// "paddle/fluid/framework/fleet/heter_ps/cudf/concurrent_unordered_map.cuh.h"
#include <iostream>
#include "paddle/fluid/framework/fleet/heter_ps/cudf/managed.cuh"
namespace
paddle
{
namespace
framework
{
class
MemoryPool
{
public:
MemoryPool
(
size_t
capacity
,
size_t
block_size
)
:
capacity_
(
capacity
),
block_size_
(
block_size
)
{
VLOG
(
3
)
<<
"mem_pool init with block_size: "
<<
block_size
<<
" capacity: "
<<
capacity
;
mem_
=
(
char
*
)
malloc
(
block_size
*
capacity_
);
}
~
MemoryPool
()
{
VLOG
(
3
)
<<
"mem pool delete"
;
free
(
mem_
);
}
size_t
block_size
()
{
return
block_size_
;
}
char
*
mem
()
{
return
mem_
;
}
size_t
capacity
()
{
return
capacity_
;
}
size_t
byte_size
()
{
return
capacity_
*
block_size_
;
}
void
*
mem_address
(
const
uint32_t
&
idx
)
{
return
(
void
*
)
&
mem_
[(
idx
)
*
block_size_
];
}
private:
char
*
mem_
=
NULL
;
size_t
capacity_
;
size_t
block_size_
;
};
class
HBMMemoryPool
:
public
managed
{
public:
HBMMemoryPool
(
size_t
capacity
,
size_t
block_size
)
:
capacity_
(
capacity
),
block_size_
(
block_size
)
{}
HBMMemoryPool
(
MemoryPool
*
mem_pool
)
{
capacity_
=
mem_pool
->
capacity
();
block_size_
=
mem_pool
->
block_size
();
VLOG
(
3
)
<<
"hbm memory pool with capacity"
<<
capacity_
<<
" bs: "
<<
block_size_
;
cudaMalloc
(
&
mem_
,
block_size_
*
capacity_
);
cudaMemcpy
(
mem_
,
mem_pool
->
mem
(),
mem_pool
->
byte_size
(),
cudaMemcpyHostToDevice
);
}
~
HBMMemoryPool
()
{
VLOG
(
3
)
<<
"delete hbm memory pool"
;
cudaFree
(
mem_
);
}
size_t
block_size
()
{
return
block_size_
;
}
void
clear
(
void
)
{
cudaMemset
(
mem_
,
0
,
block_size_
*
capacity_
);
}
void
reset
(
size_t
capacity
)
{
cudaFree
(
mem_
);
mem_
=
NULL
;
capacity_
=
capacity
;
cudaMalloc
(
&
mem_
,
(
block_size_
*
capacity
/
8
+
1
)
*
8
);
cudaMemset
(
mem_
,
0
,
block_size_
*
capacity
);
}
friend
std
::
ostream
&
operator
<<
(
std
::
ostream
&
out
,
HBMMemoryPool
&
p
)
{
for
(
size_t
k
=
0
;
k
<
5
;
k
++
)
{
auto
x
=
(
FeatureValue
*
)(
p
.
mem
()
+
k
*
p
.
capacity
());
out
<<
"show: "
<<
x
->
show
<<
" clk: "
<<
x
->
clk
<<
" slot: "
<<
x
->
slot
<<
" lr: "
<<
x
->
lr
<<
" mf_dim: "
<<
x
->
mf_size
<<
" mf_size: "
<<
x
->
mf_size
<<
" mf:"
;
for
(
int
i
=
0
;
i
<
x
->
mf_dim
+
1
;
++
i
)
{
out
<<
" "
<<
x
->
mf
[
i
];
}
out
<<
"
\n
"
;
}
return
out
;
}
char
*
mem
()
{
return
mem_
;
}
size_t
capacity
()
{
return
capacity_
;
}
__forceinline__
__device__
void
*
mem_address
(
const
uint32_t
&
idx
)
{
return
(
void
*
)
&
mem_
[(
idx
)
*
block_size_
];
}
private:
char
*
mem_
=
NULL
;
size_t
capacity_
;
size_t
block_size_
;
};
}
// end namespace framework
}
// end namespace paddle
#endif
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