std::kill_dependency

From cppreference.com
< cpp‎ | atomic
 
 
Concurrency support library
Threads
(C++11)
(C++20)
(C++20)
this_thread namespace
(C++11)
(C++11)
(C++11)
Atomic types
(C++11)
(C++20)
Initialization of atomic types
(C++11)(deprecated in C++20)
(C++11)(deprecated in C++20)
Free functions for atomic operations
Free functions for atomic flags
Memory ordering
kill_dependency
(C++11)
Mutual exclusion
(C++11)
Generic lock management
(C++11)
(C++11)
(C++11)
(C++11)(C++11)(C++11)
(C++11)
(C++11)
Condition variables
(C++11)
Semaphores
Latches and barriers
(C++20)
(C++20)
Futures
(C++11)
(C++11)
(C++11)
(C++11)
 
Defined in header <atomic>
template< class T >
T kill_dependency( T y ) noexcept;
(since C++11)

Informs the compiler that the dependency tree started by an std::memory_order_consume atomic load operation does not extend past the return value of std::kill_dependency; that is, the argument does not carry a dependency into the return value.

This may be used to avoid unnecessary std::memory_order_acquire fences when the dependency chain leaves function scope (and the function does not have the [[carries_dependency]] attribute).

Parameters

y - the expression whose return value is to be removed from a dependency tree

Return value

Returns y, no longer a part of a dependency tree.

Examples

file1.cpp:
struct Foo
{
    int* a;
    int* b;
};
 
std::atomic<Foo*> foo_head[10];
int foo_array[10][10];
 
// consume operation starts a dependency chain, which escapes this function
[[carries_dependency]] Foo* f(int i)
{
    return foo_head[i].load(memory_order_consume);
}
 
// the dependency chain enters this function through the right parameter and is
// killed before the function ends (so no extra acquire operation takes place)
int g(int* x, int* y [[carries_dependency]])
{
    return std::kill_dependency(foo_array[*x][*y]);
}
file2.cpp:
[[carries_dependency]] struct Foo* f(int i);
int g(int* x, int* y [[carries_dependency]]);
 
int c = 3;
void h(int i)
{
    Foo* p;
    p = f(i); // dependency chain started inside f continues into p without undue acquire
    do_something_with(g(&c, p->a)); // p->b is not brought in from the cache
    do_something_with(g(p->a, &c)); // left argument does not have the carries_dependency
                                    // attribute: memory acquire fence may be issued
                                    // p->b becomes visible before g() is entered
}

See also

defines memory ordering constraints for the given atomic operation
(enum)