问题描述
您能否给出一个真实的示例,其中出于某种原因使用了两个std::atomic::compare_exchange
的memory_order参数版本(因此,一个单独的memory_order参数版本不合适)?
Can you give a real-world example where two memory_order parameter version of std::atomic::compare_exchange
is used for a reason (so the one memory_order parameter version is not adequate)?
推荐答案
在许多情况下,compare_exchange
上的第二个内存排序参数设置为memory_order_relaxed
.在那些情况下,省略它通常是没有错的,只是效率可能较低.
In many cases, the second memory orderering parameter on compare_exchange
is set to memory_order_relaxed
. In those cases, it is usually not wrong to omit it, just potentially less efficient.
这里是一个简单的,无锁的列表/堆栈的示例,该列表/堆栈需要compare_exchange_weak
上的第二个不同的排序参数才能不涉及数据争用.
Here is an example of a simple, lock-free, list/stack that requires a second, different, ordering parameter on compare_exchange_weak
in order to be data-race-free.
可以同时执行对push
的调用,但是为了避免无锁数据操作的复杂性,假定执行push
的调用无法将节点从堆栈中删除;即避免悬空的指针.
Calls to push
can be executed concurrently, but to avoid the complexities of lock-free data manipulation,the assumption is made that nodes cannot be removed from the stack while calls to push
are executed; i.e. to avoid dangling pointers.
template<typename T>
class mystack {
struct node {
node *next = nullptr;
T data;
int id;
node(int id) : id{id} { }
};
std::atomic<node *> head{nullptr};
public:
void push(T data, int id);
bool pop(T &data); // not implemented
};
template<typename T>
void mystack<T>::push(T data, int id)
{
node *newnode = new node{id};
newnode->data = std::move(data);
node *current_head = head.load(std::memory_order_relaxed); // A
for (;;)
{
newnode->next = current_head;
if (head.compare_exchange_weak(current_head, newnode,
std::memory_order_release, // B
std::memory_order_acquire)) // C
{
/*
* 'current_head' may not be derefenced here since the initial load (at A)
* does not order memory 'current_head' is pointing at.
*
* a release barrier (at B) is necessary to make 'newnode' available
* to other threads
*/
std::cout << "Insertion successful\n";
break;
} else
{
/*
* 'current_head' is the updated head pointer after 'compare_exchange' failed
* Since it was inserted by another thread (the CAS failed),
* an acquire barrier must be set (at C) in order to be able to access data
* 'current_head' is pointing at.
*/
std::cout << "Insertion failed after head changed to id: " <<
current_head->id << std::endl;
}
}
}
在push
中,初始的load
(在A处)是宽松的操作,这意味着即使head
指针是原子加载的,由于它所引用的内存在此线程中是无序的,因此可能无法取消引用.
In push
, the initial load
(at A) is a relaxed operation, meaning that even though the head
pointer is loaded atomically,it may not be dereferenced since the memory it refers to is unordered in this thread.
如果compare_exchange_weak
返回成功,则通过设置释放屏障(在B处)将newnode
插入列表的开头,并使其可用于其他线程.访问此数据的另一个线程(后来通过pop
)需要设置获取屏障.
In case compare_exchange_weak
returns success, newnode
is inserted at the head of the list and made available to other threads by setting a release barrier (at B).Another thread that accesses this data (later, via pop
) needs to set an acquire barrier.
如果compare_exchange_weak
返回失败(错误地忘记),则另一个线程刚刚插入了新的node
实例,并且current_head
被新的值head
更新.由于current_head
现在指向在另一个线程中分配和释放的数据,因此如果要取消引用current_head
,则需要获取屏障.
这是正确的,因为cout
失败消息包括current_head->id
.
In case compare_exchange_weak
returns failure (forget spuriously), another thread just inserted a new node
instance and current_head
is updated with the new value of head
.Since current_head
is now pointing at data that was allocated and released in another thread, an acquire barrier is necessary if current_head
is going to be dereferenced.
This is true since the cout
failure message includes current_head->id
.
如果省略了最后一个参数,则第一个屏障参数将用于故障情况load
,但是由于这是释放屏障,有效屏障将衰减至memory_order_relaxed
,从而导致current_head->id
上的数据争用.
Had the last parameter been omitted, the first barrier parameter would have been used for the failure load
scenario, but since this is a release barrier, the effective barrier would have decayed to memory_order_relaxed
, causing a data race on current_head->id
.
这篇关于真实示例,其中std :: atomic :: compare_exchange与两个memory_order参数一起使用的文章就介绍到这了,希望我们推荐的答案对大家有所帮助,也希望大家多多支持!