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
25f8a198
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
25f8a198
编写于
8月 23, 2021
作者:
W
wanghuancoder
提交者:
GitHub
8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
Revert "use spin lock in auto growth allocator (#34910)"
This reverts commit
6bacfb0e
.
上级
56c5e210
变更
4
显示空白变更内容
内联
并排
Showing
4 changed file
with
3 addition
and
280 deletion
+3
-280
paddle/fluid/memory/allocation/auto_growth_best_fit_allocator.cc
...fluid/memory/allocation/auto_growth_best_fit_allocator.cc
+2
-3
paddle/fluid/memory/allocation/auto_growth_best_fit_allocator.h
.../fluid/memory/allocation/auto_growth_best_fit_allocator.h
+1
-2
paddle/fluid/memory/allocation/spin_lock.h
paddle/fluid/memory/allocation/spin_lock.h
+0
-36
paddle/fluid/memory/allocation/spin_lock_c.h
paddle/fluid/memory/allocation/spin_lock_c.h
+0
-239
未找到文件。
paddle/fluid/memory/allocation/auto_growth_best_fit_allocator.cc
浏览文件 @
25f8a198
...
@@ -45,7 +45,7 @@ AutoGrowthBestFitAllocator::AutoGrowthBestFitAllocator(
...
@@ -45,7 +45,7 @@ AutoGrowthBestFitAllocator::AutoGrowthBestFitAllocator(
Allocation
*
AutoGrowthBestFitAllocator
::
AllocateImpl
(
size_t
size
)
{
Allocation
*
AutoGrowthBestFitAllocator
::
AllocateImpl
(
size_t
size
)
{
size
=
AlignedSize
(
size
,
alignment_
);
size
=
AlignedSize
(
size
,
alignment_
);
std
::
lock_guard
<
SpinLock
>
guard
(
spinlock
_
);
std
::
lock_guard
<
std
::
mutex
>
guard
(
mtx
_
);
auto
iter
=
free_blocks_
.
lower_bound
(
std
::
make_pair
(
size
,
nullptr
));
auto
iter
=
free_blocks_
.
lower_bound
(
std
::
make_pair
(
size
,
nullptr
));
BlockIt
block_it
;
BlockIt
block_it
;
if
(
iter
!=
free_blocks_
.
end
())
{
if
(
iter
!=
free_blocks_
.
end
())
{
...
@@ -94,12 +94,11 @@ Allocation *AutoGrowthBestFitAllocator::AllocateImpl(size_t size) {
...
@@ -94,12 +94,11 @@ Allocation *AutoGrowthBestFitAllocator::AllocateImpl(size_t size) {
VLOG
(
2
)
<<
"Not found and reallocate "
<<
realloc_size
<<
", and remaining "
VLOG
(
2
)
<<
"Not found and reallocate "
<<
realloc_size
<<
", and remaining "
<<
remaining_size
;
<<
remaining_size
;
}
}
return
new
BlockAllocation
(
block_it
);
return
new
BlockAllocation
(
block_it
);
}
}
void
AutoGrowthBestFitAllocator
::
FreeImpl
(
Allocation
*
allocation
)
{
void
AutoGrowthBestFitAllocator
::
FreeImpl
(
Allocation
*
allocation
)
{
std
::
lock_guard
<
SpinLock
>
guard
(
spinlock
_
);
std
::
lock_guard
<
std
::
mutex
>
guard
(
mtx
_
);
auto
block_it
=
static_cast
<
BlockAllocation
*>
(
allocation
)
->
block_it_
;
auto
block_it
=
static_cast
<
BlockAllocation
*>
(
allocation
)
->
block_it_
;
auto
&
blocks
=
block_it
->
chunk_
->
blocks_
;
auto
&
blocks
=
block_it
->
chunk_
->
blocks_
;
...
...
paddle/fluid/memory/allocation/auto_growth_best_fit_allocator.h
浏览文件 @
25f8a198
...
@@ -21,7 +21,6 @@
...
@@ -21,7 +21,6 @@
#include <utility>
#include <utility>
#include "paddle/fluid/memory/allocation/allocator.h"
#include "paddle/fluid/memory/allocation/allocator.h"
#include "paddle/fluid/memory/allocation/spin_lock.h"
namespace
paddle
{
namespace
paddle
{
namespace
memory
{
namespace
memory
{
...
@@ -87,7 +86,7 @@ class AutoGrowthBestFitAllocator : public Allocator {
...
@@ -87,7 +86,7 @@ class AutoGrowthBestFitAllocator : public Allocator {
size_t
alignment_
;
size_t
alignment_
;
size_t
chunk_size_
;
size_t
chunk_size_
;
SpinLock
spinlock
_
;
mutable
std
::
mutex
mtx
_
;
};
};
}
// namespace allocation
}
// namespace allocation
...
...
paddle/fluid/memory/allocation/spin_lock.h
已删除
100644 → 0
浏览文件 @
56c5e210
// 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
#include "paddle/fluid/memory/allocation/spin_lock_c.h"
namespace
paddle
{
namespace
memory
{
class
SpinLock
{
public:
SpinLock
()
{
INITIAL_LOCK
(
&
mlock_
);
}
void
lock
()
{
ACQUIRE_LOCK
(
&
mlock_
);
}
void
unlock
()
{
RELEASE_LOCK
(
&
mlock_
);
}
DISABLE_COPY_AND_ASSIGN
(
SpinLock
);
private:
MLOCK_T
mlock_
;
};
}
// namespace memory
}
// namespace paddle
paddle/fluid/memory/allocation/spin_lock_c.h
已删除
100644 → 0
浏览文件 @
56c5e210
// 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.
/*
the spinlock implementation is borrowed from Doug Lea's malloc, released to the
public domain, as explained at
http://creativecommons.org/licenses/publicdomain. Send questions,
comments, complaints, performance data, etc to dl@cs.oswego.edu
*/
#pragma once
#ifndef WIN32
#include <pthread.h>
#if defined(__SVR4) && defined(__sun)
/* solaris */
#include <thread.h>
#endif
/* solaris */
#else
#ifndef _M_AMD64
/* These are already defined on AMD64 builds */
#ifdef __cplusplus
extern
"C"
{
#endif
/* __cplusplus */
LONG
__cdecl
_InterlockedCompareExchange
(
LONG
volatile
*
Dest
,
LONG
Exchange
,
LONG
Comp
);
LONG
__cdecl
_InterlockedExchange
(
LONG
volatile
*
Target
,
LONG
Value
);
#ifdef __cplusplus
}
#endif
/* __cplusplus */
#endif
/* _M_AMD64 */
#pragma intrinsic(_InterlockedCompareExchange)
#pragma intrinsic(_InterlockedExchange)
#define interlockedcompareexchange _InterlockedCompareExchange
#define interlockedexchange _InterlockedExchange
#endif
/* Win32 */
#ifndef FORCEINLINE
#if defined(__GNUC__)
#define FORCEINLINE __inline __attribute__((always_inline))
#elif defined(_MSC_VER)
#define FORCEINLINE __forceinline
#endif
#endif
#ifndef NOINLINE
#if defined(__GNUC__)
#define NOINLINE __attribute__((noinline))
#elif defined(_MSC_VER)
#define NOINLINE __declspec(noinline)
#else
#define NOINLINE
#endif
#endif
#ifdef __cplusplus
extern
"C"
{
#ifndef FORCEINLINE
#define FORCEINLINE inline
#endif
#endif
/* __cplusplus */
#ifndef FORCEINLINE
#define FORCEINLINE
#endif
#ifdef __cplusplus
};
/* end of extern "C" */
#endif
/* __cplusplus */
#ifndef WIN32
/* Custom pthread-style spin locks on x86 and x64 for gcc */
struct
pthread_mlock_t
{
volatile
unsigned
int
l
;
unsigned
int
c
;
pthread_t
threadid
;
};
#define MLOCK_T struct pthread_mlock_t
#define CURRENT_THREAD pthread_self()
#define INITIAL_LOCK(sl) ((sl)->threadid = 0, (sl)->l = (sl)->c = 0, 0)
#define ACQUIRE_LOCK(sl) pthread_acquire_lock(sl)
#define RELEASE_LOCK(sl) pthread_release_lock(sl)
#define TRY_LOCK(sl) pthread_try_lock(sl)
#define SPINS_PER_YIELD 63
// static MLOCK_T malloc_global_mutex = { 0, 0, 0};
static
FORCEINLINE
int
pthread_acquire_lock
(
MLOCK_T
*
sl
)
{
int
spins
=
0
;
volatile
unsigned
int
*
lp
=
&
sl
->
l
;
for
(;;)
{
if
(
*
lp
!=
0
)
{
if
(
sl
->
threadid
==
CURRENT_THREAD
)
{
++
sl
->
c
;
return
0
;
}
}
else
{
/* place args to cmpxchgl in locals to evade oddities in some gccs */
int
cmp
=
0
;
int
val
=
1
;
int
ret
;
__asm__
__volatile__
(
"lock; cmpxchgl %1, %2"
:
"=a"
(
ret
)
:
"r"
(
val
),
"m"
(
*
(
lp
)),
"0"
(
cmp
)
:
"memory"
,
"cc"
);
if
(
!
ret
)
{
assert
(
!
sl
->
threadid
);
sl
->
threadid
=
CURRENT_THREAD
;
sl
->
c
=
1
;
return
0
;
}
}
if
((
++
spins
&
SPINS_PER_YIELD
)
==
0
)
{
#if defined(__SVR4) && defined(__sun)
/* solaris */
thr_yield
();
#else
#if defined(__linux__) || defined(__FreeBSD__) || defined(__APPLE__)
sched_yield
();
#else
/* no-op yield on unknown systems */
;
// NOLINT
#endif
/* __linux__ || __FreeBSD__ || __APPLE__ */
#endif
/* solaris */
}
}
}
static
FORCEINLINE
void
pthread_release_lock
(
MLOCK_T
*
sl
)
{
volatile
unsigned
int
*
lp
=
&
sl
->
l
;
assert
(
*
lp
!=
0
);
assert
(
sl
->
threadid
==
CURRENT_THREAD
);
if
(
--
sl
->
c
==
0
)
{
sl
->
threadid
=
0
;
int
prev
=
0
;
int
ret
;
__asm__
__volatile__
(
"lock; xchgl %0, %1"
:
"=r"
(
ret
)
:
"m"
(
*
(
lp
)),
"0"
(
prev
)
:
"memory"
);
}
}
static
FORCEINLINE
int
pthread_try_lock
(
MLOCK_T
*
sl
)
{
volatile
unsigned
int
*
lp
=
&
sl
->
l
;
if
(
*
lp
!=
0
)
{
if
(
sl
->
threadid
==
CURRENT_THREAD
)
{
++
sl
->
c
;
return
1
;
}
}
else
{
int
cmp
=
0
;
int
val
=
1
;
int
ret
;
__asm__
__volatile__
(
"lock; cmpxchgl %1, %2"
:
"=a"
(
ret
)
:
"r"
(
val
),
"m"
(
*
(
lp
)),
"0"
(
cmp
)
:
"memory"
,
"cc"
);
if
(
!
ret
)
{
assert
(
!
sl
->
threadid
);
sl
->
threadid
=
CURRENT_THREAD
;
sl
->
c
=
1
;
return
1
;
}
}
return
0
;
}
#else
/* WIN32 */
/* Custom win32-style spin locks on x86 and x64 for MSC */
struct
win32_mlock_t
{
volatile
long
l
;
// NOLINT
unsigned
int
c
;
long
threadid
;
// NOLINT
};
#define MLOCK_T struct win32_mlock_t
#define CURRENT_THREAD GetCurrentThreadId()
#define INITIAL_LOCK(sl) ((sl)->threadid = 0, (sl)->l = (sl)->c = 0, 0)
#define ACQUIRE_LOCK(sl) win32_acquire_lock(sl)
#define RELEASE_LOCK(sl) win32_release_lock(sl)
#define TRY_LOCK(sl) win32_try_lock(sl)
#define SPINS_PER_YIELD 63
// static MLOCK_T malloc_global_mutex = { 0, 0, 0};
static
FORCEINLINE
int
win32_acquire_lock
(
MLOCK_T
*
sl
)
{
int
spins
=
0
;
for
(;;)
{
if
(
sl
->
l
!=
0
)
{
if
(
sl
->
threadid
==
CURRENT_THREAD
)
{
++
sl
->
c
;
return
0
;
}
}
else
{
if
(
!
interlockedexchange
(
&
sl
->
l
,
1
))
{
assert
(
!
sl
->
threadid
);
sl
->
threadid
=
CURRENT_THREAD
;
sl
->
c
=
1
;
return
0
;
}
}
if
((
++
spins
&
SPINS_PER_YIELD
)
==
0
)
SleepEx
(
0
,
FALSE
);
}
}
static
FORCEINLINE
void
win32_release_lock
(
MLOCK_T
*
sl
)
{
assert
(
sl
->
threadid
==
CURRENT_THREAD
);
assert
(
sl
->
l
!=
0
);
if
(
--
sl
->
c
==
0
)
{
sl
->
threadid
=
0
;
interlockedexchange
(
&
sl
->
l
,
0
);
}
}
static
FORCEINLINE
int
win32_try_lock
(
MLOCK_T
*
sl
)
{
if
(
sl
->
l
!=
0
)
{
if
(
sl
->
threadid
==
CURRENT_THREAD
)
{
++
sl
->
c
;
return
1
;
}
}
else
{
if
(
!
interlockedexchange
(
&
sl
->
l
,
1
))
{
assert
(
!
sl
->
threadid
);
sl
->
threadid
=
CURRENT_THREAD
;
sl
->
c
=
1
;
return
1
;
}
}
return
0
;
}
#endif
/* WIN32 */
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