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
318072c2
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
提交
318072c2
编写于
3月 26, 2019
作者:
S
sneaxiy
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
add comments of allocator design
test=develop
上级
2d92b6be
变更
1
显示空白变更内容
内联
并排
Showing
1 changed file
with
39 addition
and
0 deletion
+39
-0
paddle/fluid/memory/allocation/allocator.h
paddle/fluid/memory/allocation/allocator.h
+39
-0
未找到文件。
paddle/fluid/memory/allocation/allocator.h
浏览文件 @
318072c2
...
...
@@ -46,10 +46,49 @@ class Allocator;
// NOTE: this is the base class of Allocation. Each allocator can use its own
// allocation object.
// NOTE: the `Allocation::ptr()` could be nullptr, if the allocation size is 0
/**
* Allocation is returned by Allocator::Allocate() method.
*
* An allocator may be decorated by another allocator. For example, we can
* decorate
* a RetryAllocator to any allocator to perform allocation retry when first
* allocation request fails.
*
* Explanations of Allocator design is as follows:
*
* Suppose we have an allocator which is decorated by several allocators:
*
* A(1) <- A(2) <- A(3) <- ... <- A(n)
*
* , and the public allocator is A(1).
*
* The allocation process would be:
*
* A(n).Allocate() -> ... -> A(2).Allocate() -> A(1).Allocate()
*
* , and the free process would be:
*
* A(1).Free() -> A(2).Free() -> ... -> A(n).Free()
*
* Therefore, we should record the allocator chain when allocating, so
* that we can free the allocation in the reverse order of allocator chain.
* The field `decorated_allocators_` is used to record this chain.
*
* Another example is that we want to add additional fields in Allocation,
* e.g., something what is done in AlignedAllocator, etc.
* In this case, we should declare a derived class of Allocation, which
* contains an underlying Allocation allocated by the underlying allocator.
* Therefore, `decorated_allocators_` of the new Allocation object would
* be a new chain, differing from the underlying Allocation object.
*/
class
Allocation
{
public:
Allocation
(
void
*
ptr
,
size_t
size
,
platform
::
Place
place
)
:
ptr_
(
ptr
),
size_
(
size
),
place_
(
place
)
{
// NOTE(zjl): Since decorated_allocators_ is usually a small vector
// We reserve a small buffer to it to prevent frequent heap allocation
// Not quite sure whether we need something like gtl vector.
decorated_allocators_
.
reserve
(
8
);
}
...
...
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