Struct Atomic

pub struct Atomic<T: ?Sized + Pointable> { /* private fields */ }

An atomic pointer that can be safely shared between threads.

The pointer must be properly aligned. Since it is aligned, a tag can be stored into the unused least significant bits of the address. For example, the tag for a pointer to a sized type T should be less than (1 << mem::align_of::<T>().trailing_zeros()).

Any method that loads the pointer must be passed a reference to a Guard.

Crossbeam supports dynamically sized types. See Pointable for details.

Implementations

impl<T> Atomic<T>

pub fn new(init: T) -> Atomic<T>

Allocates value on the heap and returns a new atomic pointer pointing to it.

Examples

use crossbeam_epoch::Atomic;

let a = Atomic::new(1234);

impl<T: ?Sized + Pointable> Atomic<T>

pub fn init(init: T::Init) -> Atomic<T>

Allocates value on the heap and returns a new atomic pointer pointing to it.

Examples

use crossbeam_epoch::Atomic;

let a = Atomic::<i32>::init(1234);

pub const fn null() -> Atomic<T>

Returns a new null atomic pointer.

Examples

use crossbeam_epoch::Atomic;

let a = Atomic::<i32>::null();

pub fn load<'g>(&self, ord: Ordering, _: &'g Guard) -> Shared<'g, T>

Loads a Shared from the atomic pointer.

This method takes an Ordering argument which describes the memory ordering of this operation.

Examples

use crossbeam_epoch::{self as epoch, Atomic};
use std::sync::atomic::Ordering::SeqCst;

let a = Atomic::new(1234);
let guard = &epoch::pin();
let p = a.load(SeqCst, guard);

pub fn load_consume<'g>(&self, _: &'g Guard) -> Shared<'g, T>

Loads a Shared from the atomic pointer using a “consume” memory ordering.

This is similar to the “acquire” ordering, except that an ordering is only guaranteed with operations that “depend on” the result of the load. However consume loads are usually much faster than acquire loads on architectures with a weak memory model since they don’t require memory fence instructions.

The exact definition of “depend on” is a bit vague, but it works as you would expect in practice since a lot of software, especially the Linux kernel, rely on this behavior.

Examples

use crossbeam_epoch::{self as epoch, Atomic};

let a = Atomic::new(1234);
let guard = &epoch::pin();
let p = a.load_consume(guard);

pub fn store<P: Pointer<T>>(&self, new: P, ord: Ordering)

Stores a Shared or Owned pointer into the atomic pointer.

This method takes an Ordering argument which describes the memory ordering of this operation.

Examples

use crossbeam_epoch::{Atomic, Owned, Shared};
use std::sync::atomic::Ordering::SeqCst;

let a = Atomic::new(1234);
a.store(Shared::null(), SeqCst);
a.store(Owned::new(1234), SeqCst);

pub fn swap<'g, P: Pointer<T>>( &self, new: P, ord: Ordering, _: &'g Guard, ) -> Shared<'g, T>

Stores a Shared or Owned pointer into the atomic pointer, returning the previous Shared.

This method takes an Ordering argument which describes the memory ordering of this operation.

Examples

use crossbeam_epoch::{self as epoch, Atomic, Shared};
use std::sync::atomic::Ordering::SeqCst;

let a = Atomic::new(1234);
let guard = &epoch::pin();
let p = a.swap(Shared::null(), SeqCst, guard);

pub fn compare_exchange<'g, P>( &self, current: Shared<'_, T>, new: P, success: Ordering, failure: Ordering, _: &'g Guard, ) -> Result<Shared<'g, T>, CompareExchangeError<'g, T, P>>

where P: Pointer<T>,

Stores the pointer new (either Shared or Owned) into the atomic pointer if the current value is the same as current. The tag is also taken into account, so two pointers to the same object, but with different tags, will not be considered equal.

The return value is a result indicating whether the new pointer was written. On success the pointer that was written is returned. On failure the actual current value and new are returned.

This method takes two Ordering arguments to describe the memory ordering of this operation. success describes the required ordering for the read-modify-write operation that takes place if the comparison with current succeeds. failure describes the required ordering for the load operation that takes place when the comparison fails. Using Acquire as success ordering makes the store part of this operation Relaxed, and using Release makes the successful load Relaxed. The failure ordering can only be SeqCst, Acquire or Relaxed and must be equivalent to or weaker than the success ordering.

Examples

use crossbeam_epoch::{self as epoch, Atomic, Owned, Shared};
use std::sync::atomic::Ordering::SeqCst;

let a = Atomic::new(1234);

let guard = &epoch::pin();
let curr = a.load(SeqCst, guard);
let res1 = a.compare_exchange(curr, Shared::null(), SeqCst, SeqCst, guard);
let res2 = a.compare_exchange(curr, Owned::new(5678), SeqCst, SeqCst, guard);

pub fn compare_exchange_weak<'g, P>( &self, current: Shared<'_, T>, new: P, success: Ordering, failure: Ordering, _: &'g Guard, ) -> Result<Shared<'g, T>, CompareExchangeError<'g, T, P>>

where P: Pointer<T>,

Stores the pointer new (either Shared or Owned) into the atomic pointer if the current value is the same as current. The tag is also taken into account, so two pointers to the same object, but with different tags, will not be considered equal.

Unlike compare_exchange, this method is allowed to spuriously fail even when comparison succeeds, which can result in more efficient code on some platforms. The return value is a result indicating whether the new pointer was written. On success the pointer that was written is returned. On failure the actual current value and new are returned.

This method takes two Ordering arguments to describe the memory ordering of this operation. success describes the required ordering for the read-modify-write operation that takes place if the comparison with current succeeds. failure describes the required ordering for the load operation that takes place when the comparison fails. Using Acquire as success ordering makes the store part of this operation Relaxed, and using Release makes the successful load Relaxed. The failure ordering can only be SeqCst, Acquire or Relaxed and must be equivalent to or weaker than the success ordering.

Examples

use crossbeam_epoch::{self as epoch, Atomic, Owned, Shared};
use std::sync::atomic::Ordering::SeqCst;

let a = Atomic::new(1234);
let guard = &epoch::pin();

let mut new = Owned::new(5678);
let mut ptr = a.load(SeqCst, guard);
loop {
    match a.compare_exchange_weak(ptr, new, SeqCst, SeqCst, guard) {
        Ok(p) => {
            ptr = p;
            break;
        }
        Err(err) => {
            ptr = err.current;
            new = err.new;
        }
    }
}

let mut curr = a.load(SeqCst, guard);
loop {
    match a.compare_exchange_weak(curr, Shared::null(), SeqCst, SeqCst, guard) {
        Ok(_) => break,
        Err(err) => curr = err.current,
    }
}

pub fn fetch_update<'g, F>( &self, set_order: Ordering, fail_order: Ordering, guard: &'g Guard, func: F, ) -> Result<Shared<'g, T>, Shared<'g, T>>

where F: FnMut(Shared<'g, T>) -> Option<Shared<'g, T>>,

Fetches the pointer, and then applies a function to it that returns a new value. Returns a Result of Ok(previous_value) if the function returned Some, else Err(_).

Note that the given function may be called multiple times if the value has been changed by other threads in the meantime, as long as the function returns Some(_), but the function will have been applied only once to the stored value.

fetch_update takes two Ordering arguments to describe the memory ordering of this operation. The first describes the required ordering for when the operation finally succeeds while the second describes the required ordering for loads. These correspond to the success and failure orderings of Atomic::compare_exchange respectively.

Using Acquire as success ordering makes the store part of this operation Relaxed, and using Release makes the final successful load Relaxed. The (failed) load ordering can only be SeqCst, Acquire or Relaxed and must be equivalent to or weaker than the success ordering.

Examples

use crossbeam_epoch::{self as epoch, Atomic};
use std::sync::atomic::Ordering::SeqCst;

let a = Atomic::new(1234);
let guard = &epoch::pin();

let res1 = a.fetch_update(SeqCst, SeqCst, guard, |x| Some(x.with_tag(1)));
assert!(res1.is_ok());

let res2 = a.fetch_update(SeqCst, SeqCst, guard, |x| None);
assert!(res2.is_err());

pub fn compare_and_set<'g, O, P>( &self, current: Shared<'_, T>, new: P, ord: O, guard: &'g Guard, ) -> Result<Shared<'g, T>, CompareAndSetError<'g, T, P>>

where O: CompareAndSetOrdering, P: Pointer<T>,

👎Deprecated: Use compare_exchange instead

Stores the pointer new (either Shared or Owned) into the atomic pointer if the current value is the same as current. The tag is also taken into account, so two pointers to the same object, but with different tags, will not be considered equal.

The return value is a result indicating whether the new pointer was written. On success the pointer that was written is returned. On failure the actual current value and new are returned.

This method takes a CompareAndSetOrdering argument which describes the memory ordering of this operation.

Migrating to compare_exchange

compare_and_set is equivalent to compare_exchange with the following mapping for memory orderings:

Original	Success	Failure
Relaxed	Relaxed	Relaxed
Acquire	Acquire	Acquire
Release	Release	Relaxed
AcqRel	AcqRel	Acquire
SeqCst	SeqCst	SeqCst

Examples

use crossbeam_epoch::{self as epoch, Atomic, Owned, Shared};
use std::sync::atomic::Ordering::SeqCst;

let a = Atomic::new(1234);

let guard = &epoch::pin();
let curr = a.load(SeqCst, guard);
let res1 = a.compare_and_set(curr, Shared::null(), SeqCst, guard);
let res2 = a.compare_and_set(curr, Owned::new(5678), SeqCst, guard);

pub fn compare_and_set_weak<'g, O, P>( &self, current: Shared<'_, T>, new: P, ord: O, guard: &'g Guard, ) -> Result<Shared<'g, T>, CompareAndSetError<'g, T, P>>

where O: CompareAndSetOrdering, P: Pointer<T>,

👎Deprecated: Use compare_exchange_weak instead

Stores the pointer new (either Shared or Owned) into the atomic pointer if the current value is the same as current. The tag is also taken into account, so two pointers to the same object, but with different tags, will not be considered equal.

Unlike compare_and_set, this method is allowed to spuriously fail even when comparison succeeds, which can result in more efficient code on some platforms. The return value is a result indicating whether the new pointer was written. On success the pointer that was written is returned. On failure the actual current value and new are returned.

This method takes a CompareAndSetOrdering argument which describes the memory ordering of this operation.

Migrating to compare_exchange_weak

compare_and_set_weak is equivalent to compare_exchange_weak with the following mapping for memory orderings:

Original	Success	Failure
Relaxed	Relaxed	Relaxed
Acquire	Acquire	Acquire
Release	Release	Relaxed
AcqRel	AcqRel	Acquire
SeqCst	SeqCst	SeqCst

Examples

use crossbeam_epoch::{self as epoch, Atomic, Owned, Shared};
use std::sync::atomic::Ordering::SeqCst;

let a = Atomic::new(1234);
let guard = &epoch::pin();

let mut new = Owned::new(5678);
let mut ptr = a.load(SeqCst, guard);
loop {
    match a.compare_and_set_weak(ptr, new, SeqCst, guard) {
        Ok(p) => {
            ptr = p;
            break;
        }
        Err(err) => {
            ptr = err.current;
            new = err.new;
        }
    }
}

let mut curr = a.load(SeqCst, guard);
loop {
    match a.compare_and_set_weak(curr, Shared::null(), SeqCst, guard) {
        Ok(_) => break,
        Err(err) => curr = err.current,
    }
}

pub fn fetch_and<'g>( &self, val: usize, ord: Ordering, _: &'g Guard, ) -> Shared<'g, T>

Bitwise “and” with the current tag.

Performs a bitwise “and” operation on the current tag and the argument val, and sets the new tag to the result. Returns the previous pointer.

This method takes an Ordering argument which describes the memory ordering of this operation.

Examples

use crossbeam_epoch::{self as epoch, Atomic, Shared};
use std::sync::atomic::Ordering::SeqCst;

let a = Atomic::<i32>::from(Shared::null().with_tag(3));
let guard = &epoch::pin();
assert_eq!(a.fetch_and(2, SeqCst, guard).tag(), 3);
assert_eq!(a.load(SeqCst, guard).tag(), 2);

pub fn fetch_or<'g>( &self, val: usize, ord: Ordering, _: &'g Guard, ) -> Shared<'g, T>

Bitwise “or” with the current tag.

Performs a bitwise “or” operation on the current tag and the argument val, and sets the new tag to the result. Returns the previous pointer.

This method takes an Ordering argument which describes the memory ordering of this operation.

Examples

use crossbeam_epoch::{self as epoch, Atomic, Shared};
use std::sync::atomic::Ordering::SeqCst;

let a = Atomic::<i32>::from(Shared::null().with_tag(1));
let guard = &epoch::pin();
assert_eq!(a.fetch_or(2, SeqCst, guard).tag(), 1);
assert_eq!(a.load(SeqCst, guard).tag(), 3);

pub fn fetch_xor<'g>( &self, val: usize, ord: Ordering, _: &'g Guard, ) -> Shared<'g, T>

Bitwise “xor” with the current tag.

Performs a bitwise “xor” operation on the current tag and the argument val, and sets the new tag to the result. Returns the previous pointer.

This method takes an Ordering argument which describes the memory ordering of this operation.

Examples

use crossbeam_epoch::{self as epoch, Atomic, Shared};
use std::sync::atomic::Ordering::SeqCst;

let a = Atomic::<i32>::from(Shared::null().with_tag(1));
let guard = &epoch::pin();
assert_eq!(a.fetch_xor(3, SeqCst, guard).tag(), 1);
assert_eq!(a.load(SeqCst, guard).tag(), 2);

pub unsafe fn into_owned(self) -> Owned<T>

Takes ownership of the pointee.

This consumes the atomic and converts it into Owned. As Atomic doesn’t have a destructor and doesn’t drop the pointee while Owned does, this is suitable for destructors of data structures.

Panics

Panics if this pointer is null, but only in debug mode.

Safety

This method may be called only if the pointer is valid and nobody else is holding a reference to the same object.

Examples

struct DataStructure {
    ptr: Atomic<usize>,
}

impl Drop for DataStructure {
    fn drop(&mut self) {
        // By now the DataStructure lives only in our thread and we are sure we don't hold
        // any Shared or & to it ourselves.
        unsafe {
            drop(mem::replace(&mut self.ptr, Atomic::null()).into_owned());
        }
    }
}

pub unsafe fn try_into_owned(self) -> Option<Owned<T>>

Takes ownership of the pointee if it is non-null.

This consumes the atomic and converts it into Owned. As Atomic doesn’t have a destructor and doesn’t drop the pointee while Owned does, this is suitable for destructors of data structures.

Safety

This method may be called only if the pointer is valid and nobody else is holding a reference to the same object, or the pointer is null.

Examples

struct DataStructure {
    ptr: Atomic<usize>,
}

impl Drop for DataStructure {
    fn drop(&mut self) {
        // By now the DataStructure lives only in our thread and we are sure we don't hold
        // any Shared or & to it ourselves, but it may be null, so we have to be careful.
        let old = mem::replace(&mut self.ptr, Atomic::null());
        unsafe {
            if let Some(x) = old.try_into_owned() {
                drop(x)
            }
        }
    }
}

Trait Implementations

impl<T: ?Sized + Pointable> Clone for Atomic<T>

fn clone(&self) -> Self

Returns a copy of the atomic value.

Note that a Relaxed load is used here. If you need synchronization, use it with other atomics or fences.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<T: ?Sized + Pointable> Debug for Atomic<T>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<T: ?Sized + Pointable> Default for Atomic<T>

fn default() -> Self

Returns the “default value” for a type.

impl<T> From<*const T> for Atomic<T>

fn from(raw: *const T) -> Self

Returns a new atomic pointer pointing to raw.

Examples

use std::ptr;
use crossbeam_epoch::Atomic;

let a = Atomic::<i32>::from(ptr::null::<i32>());

impl<T> From<Box<T>> for Atomic<T>

fn from(b: Box<T>) -> Self

Converts to this type from the input type.

impl<T: ?Sized + Pointable> From<Owned<T>> for Atomic<T>

fn from(owned: Owned<T>) -> Self

Returns a new atomic pointer pointing to owned.

Examples

use crossbeam_epoch::{Atomic, Owned};

let a = Atomic::<i32>::from(Owned::new(1234));

impl<'g, T: ?Sized + Pointable> From<Shared<'g, T>> for Atomic<T>

fn from(ptr: Shared<'g, T>) -> Self

Returns a new atomic pointer pointing to ptr.

Examples

use crossbeam_epoch::{Atomic, Shared};

let a = Atomic::<i32>::from(Shared::<i32>::null());

impl<T> From<T> for Atomic<T>

fn from(t: T) -> Self

Converts to this type from the input type.

impl<T: ?Sized + Pointable> Pointer for Atomic<T>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<T: ?Sized + Pointable + Send + Sync> Send for Atomic<T>

impl<T: ?Sized + Pointable + Send + Sync> Sync for Atomic<T>

Layout

Note: Most layout information is completely unstable and may even differ between compilations. The only exception is types with certain repr(...) attributes. Please see the Rust Reference's “Type Layout” chapter for details on type layout guarantees.

Size: 8 bytes