classicvalues / swift-atomics

Low-level atomic operations for Swift

Swift Atomics ⚛︎︎

This package implements an atomics library for Swift, providing atomic operations for a variety of Swift types, including integers and pointer values. The goal is to enable intrepid developers to start building synchronization constructs directly in Swift.

Atomic operations aren't subject to the usual exclusivity rules. The same memory location may be safely read and updated from multiple concurrent threads of execution, as long as all such access is done through atomic operations. For example, here is a trivial atomic counter:

import Atomics
import Dispatch

let counter = ManagedAtomic<Int>(0)

DispatchQueue.concurrentPerform(iterations: 10) { _ in
  for _ in 0 ..< 1_000_000 {
    counter.wrappingIncrement(ordering: .relaxed)
  }
}
counter.load(ordering: .relaxed) // ⟹ 10_000_000

The only way to access the counter value is to use one of the methods provided by ManagedAtomic, each of which implement a particular atomic operation, and each of which require an explicit ordering value. (Swift supports a subset of the C/C++ memory orderings.)

Table of Contents

• Getting Started
• Features
  • Lock-Free vs Wait-Free Operations
  • Portability Concerns
  • Memory Management
• Source Stability
• Contributing to Swift Atomics
• Development

Getting Started

To use Atomics in your own project, you need to set it up as a package dependency:

// swift-tools-version:5.3
import PackageDescription

let package = Package(
  name: "MyPackage",
  dependencies: [
    .package(
      url: "https://github.com/apple/swift-atomics.git", 
      .upToNextMajor(from: "1.0.0") // or `.upToNextMinor
    )
  ],
  targets: [
    .target(
      name: "MyTarget",
      dependencies: [
        .product(name: "Atomics", package: "swift-atomics")
      ]
    )
  ]
)

Features

The package implements atomic operations for the following Swift constructs, all of which conform to the public AtomicValue protocol:

• Standard signed integer types (Int, Int64, Int32, Int16, Int8)
• Standard unsigned integer types (UInt, UInt64, UInt32, UInt16, UInt8)
• Booleans (Bool)
• Standard pointer types (UnsafeRawPointer, UnsafeMutableRawPointer, UnsafePointer<T>, UnsafeMutablePointer<T>), along with their optional-wrapped forms (such as Optional<UnsafePointer<T>>)
• Unmanaged references (Unmanaged<T>, Optional<Unmanaged<T>>)
• A special DoubleWord type that consists of two UInt values, low and high, providing double-wide atomic primitives
• Any RawRepresentable type whose RawValue is in turn an atomic type (such as simple custom enum types)
• Strong references to class instances that opted into atomic use (by conforming to the AtomicReference protocol)

Of particular note is full support for atomic strong references. This provides a convenient memory reclamation solution for concurrent data structures that fits perfectly with Swift's reference counting memory management model. (Atomic strong references are implemented in terms of DoubleWord operations.) However, accessing an atomic strong reference is (relatively) expensive, so we also provide a separate set of efficient constructs (ManagedAtomicLazyReference and UnsafeAtomicLazyReference) for the common case of a lazily initialized (but otherwise constant) atomic strong reference.

Lock-Free vs Wait-Free Operations

All atomic operations exposed by this package are guaranteed to have lock-free implementations. However, we do not guarantee wait-free operation -- depending on the capabilities of the target platform, some of the exposed operations may be implemented by compare-and-exchange loops. That said, all atomic operations map directly to dedicated CPU instructions where available -- to the extent supported by llvm & Clang.

Portability Concerns

Lock-free double-wide atomics requires support for such things from the underlying target platform. Where such support isn't available, this package doesn't implement DoubleWord atomics or atomic strong references. While modern multiprocessing CPUs have been providing double-wide atomic instructions for a number of years now, some platforms still target older architectures by default; these require a special compiler option to enable double-wide atomic instructions. This currently includes Linux operating systems running on x86_64 processors, where the cmpxchg16b instruction isn't considered a baseline requirement.

To enable double-wide atomics on Linux/x86_64, you currently have to manually supply a couple of additional options on the SPM build invocation:

$ swift build -Xcc -mcx16 -Xswiftc -DENABLE_DOUBLEWIDE_ATOMICS -c release

(-mcx16 turns on support for cmpxchg16b in Clang, and -DENABLE_DOUBLEWIDE_ATOMICS makes Swift aware that double-wide atomics are available. Note that the resulting binaries won't run on some older AMD64 CPUs.)

The package cannot currently configure this automatically.

Memory Management

Atomic access is implemented in terms of dedicated atomic storage representations that are kept distinct from the corresponding regular (non-atomic) type. (E.g., the actual integer value underlying the counter above isn't directly accessible.) This has several advantages:

• it helps prevent accidental non-atomic access to atomic variables,
• it enables custom storage representations (such as the one used by atomic strong references), and
• it is a better fit with the standard C atomics library that we use to implement the actual operations (as enabled by SE-0282).

While the underlying pointer-based atomic operations are exposed as static methods on the corresponding AtomicStorage types, we strongly recommend the use of higher-level atomic wrappers to manage the details of preparing/disposing atomic storage. This version of the library provides two wrapper types:

• an easy to use, memory-safe ManagedAtomic<T> generic class and
• a less convenient, but more flexible UnsafeAtomic<T> generic struct.

Both constructs provide the following operations on all AtomicValue types:

func load(ordering: AtomicLoadOrdering) -> Value
func store(_ desired: Value, ordering: AtomicStoreOrdering)
func exchange(_ desired: Value, ordering: AtomicUpdateOrdering) -> Value

func compareExchange(
    expected: Value,
    desired: Value,
    ordering: AtomicUpdateOrdering
) -> (exchanged: Bool, original: Value)

func compareExchange(
    expected: Value,
    desired: Value,
    successOrdering: AtomicUpdateOrdering,
    failureOrdering: AtomicLoadOrdering
) -> (exchanged: Bool, original: Value)

func weakCompareExchange(
    expected: Value,
    desired: Value,
    successOrdering: AtomicUpdateOrdering,
    failureOrdering: AtomicLoadOrdering
) -> (exchanged: Bool, original: Value)

Integer types come with additional atomic operations for incrementing or decrementing values and bitwise logical operations. Bool provides select additional boolean operations along the same vein.

For an introduction to the APIs provided by this package, for now please see the first version of SE-0282.

Note that when/if Swift gains support for non-copiable types, we expect to replace both ManagedAtomic and UnsafeAtomic with a single move-only atomic struct that combines the performance and versatility of UnsafeAtomic with the ease-of-use and memory safety of ManagedAtomic.

The current version of the Atomics module does not implement APIs for tagged atomics (see issue #1), although it does expose a DoubleWord type that can be used to implement them. (Atomic strong references are already implemented in terms of DoubleWord, although in their current form they do not expose any user-customizable bits.)

Source Stability

The Swift Atomics package is source stable. The version numbers follow Semantic Versioning -- source breaking changes to public API can only land in a new major version.

The public API of version 1.0 of the swift-atomics package consists of non-underscored declarations that are marked public in the Atomics module.

By "underscored declarations" we mean declarations that have a leading underscore anywhere in their fully qualified name. For instance, here are some names that wouldn't be considered part of the public API, even if they were technically marked public:

• FooModule.Bar._someMember(value:) (underscored member)
• FooModule._Bar.someMember (underscored type)
• _FooModule.Bar (underscored module)
• FooModule.Bar.init(_value:) (underscored initializer)

Interfaces that aren't part of the public API may continue to change in any release, including patch releases.

Note that contents of the _AtomicsShims module explicitly aren't public API. (As implied by its underscored module name.) The definitions therein may therefore change at whim, and the entire module may be removed in any new release -- do not import this module directly. We also don't make any source compatibility promises about the contents of the Utilities and Tests subdirectories.

If you have a use case that requires using underscored APIs, please submit a Feature Request describing it! We'd like the public interface to be as useful as possible -- although preferably without compromising safety or limiting future evolution.

Future minor versions of the package may introduce changes to these rules as needed.

We'd like this package to quickly embrace Swift language and toolchain improvements that are relevant to its mandate. Accordingly, from time to time, we expect that new versions of this package will require clients to upgrade to a more recent Swift toolchain release. (This allows the package to make use of new language/stdlib features, build on compiler bug fixes, and adopt new package manager functionality as soon as they are available.)

Requiring a new Swift release will only require a minor version bump.

Contributing to Swift Atomics

Swift Atomics is a standalone library separate from the core Swift project. We expect some of the atomics APIs may eventually get incorporated into the Swift Standard Library. If and when that happens such changes will be proposed to the Swift Standard Library using the established evolution process of the Swift project.

This library is licensed under the Swift License. For more information, see the Swift.org Community Guidelines, Contribution Guidelines, as well as the files LICENSE.txt, CONTRIBUTING.md and CODE_OF_CONDUCT.md at the root of this repository.

Swift Atomics uses GitHub issues to track bugs and enhancement requests. We use pull requests for development.

We have a dedicated Swift Atomics Forum where people can ask and answer questions on how to use or work on this package. It's also a great place to discuss its evolution.

If you find something that looks like a bug, please open a Bug Report! Fill out as many details as you can.

To fix a small issue or make a tiny improvement, simply submit a PR with the changes you want to make. If there is an existing issue for the bug you're fixing, please include a reference to it. Make sure to add tests covering whatever changes you are making.

For larger feature additions, it's a good idea to discuss your idea in a new Feature Request or on the forum before starting to work on it. If the discussions indicate the feature would be desirable, submit the implementation in a PR, and participate in its review discussion.

Development

This package defines a large number of similar-but-not-quite-the-same operations. To make it easier to maintain these, we use code generation to produce them.

A number of source files have a .swift.gyb extension. These are using a Python-based code generation utility called gyb which we also use within the Swift Standard Library (the name is short for Generate Your Boilerplate). To make sure the package remains buildable by SPM, the autogenerated output files are committed into this repository. You should not edit the contents of autogenerated subdirectories, or your changes will get overwritten the next time the code is regenerated.

To regenerate sources (and to update the inventory of XCTest tests), you need to manually run the script generate-sources.sh in the Utilities folder of this repository. This needs to be done every time you modify one of the template files.

The same script also runs swift test --generate-linuxmain to register any newly added unit tests.

In addition to gyb, the _AtomicsShims.h header file uses the C preprocessor to define trivial wrapper functions for every supported atomic operation -- memory ordering pairing.

⚛︎︎