1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
#![stable(feature = "core_hint", since = "1.27.0")]
//! Hints to compiler that affects how code should be emitted or optimized.
//! Hints may be compile time or runtime.
use crate::intrinsics;
use crate::ub_checks;
/// Informs the compiler that the site which is calling this function is not
/// reachable, possibly enabling further optimizations.
///
/// # Safety
///
/// Reaching this function is *Undefined Behavior*.
///
/// As the compiler assumes that all forms of Undefined Behavior can never
/// happen, it will eliminate all branches in the surrounding code that it can
/// determine will invariably lead to a call to `unreachable_unchecked()`.
///
/// If the assumptions embedded in using this function turn out to be wrong -
/// that is, if the site which is calling `unreachable_unchecked()` is actually
/// reachable at runtime - the compiler may have generated nonsensical machine
/// instructions for this situation, including in seemingly unrelated code,
/// causing difficult-to-debug problems.
///
/// Use this function sparingly. Consider using the [`unreachable!`] macro,
/// which may prevent some optimizations but will safely panic in case it is
/// actually reached at runtime. Benchmark your code to find out if using
/// `unreachable_unchecked()` comes with a performance benefit.
///
/// # Examples
///
/// `unreachable_unchecked()` can be used in situations where the compiler
/// can't prove invariants that were previously established. Such situations
/// have a higher chance of occurring if those invariants are upheld by
/// external code that the compiler can't analyze.
/// ```
/// fn prepare_inputs(divisors: &mut Vec<u32>) {
/// // Note to future-self when making changes: The invariant established
/// // here is NOT checked in `do_computation()`; if this changes, you HAVE
/// // to change `do_computation()`.
/// divisors.retain(|divisor| *divisor != 0)
/// }
///
/// /// # Safety
/// /// All elements of `divisor` must be non-zero.
/// unsafe fn do_computation(i: u32, divisors: &[u32]) -> u32 {
/// divisors.iter().fold(i, |acc, divisor| {
/// // Convince the compiler that a division by zero can't happen here
/// // and a check is not needed below.
/// if *divisor == 0 {
/// // Safety: `divisor` can't be zero because of `prepare_inputs`,
/// // but the compiler does not know about this. We *promise*
/// // that we always call `prepare_inputs`.
/// std::hint::unreachable_unchecked()
/// }
/// // The compiler would normally introduce a check here that prevents
/// // a division by zero. However, if `divisor` was zero, the branch
/// // above would reach what we explicitly marked as unreachable.
/// // The compiler concludes that `divisor` can't be zero at this point
/// // and removes the - now proven useless - check.
/// acc / divisor
/// })
/// }
///
/// let mut divisors = vec![2, 0, 4];
/// prepare_inputs(&mut divisors);
/// let result = unsafe {
/// // Safety: prepare_inputs() guarantees that divisors is non-zero
/// do_computation(100, &divisors)
/// };
/// assert_eq!(result, 12);
///
/// ```
///
/// While using `unreachable_unchecked()` is perfectly sound in the following
/// example, as the compiler is able to prove that a division by zero is not
/// possible, benchmarking reveals that `unreachable_unchecked()` provides
/// no benefit over using [`unreachable!`], while the latter does not introduce
/// the possibility of Undefined Behavior.
///
/// ```
/// fn div_1(a: u32, b: u32) -> u32 {
/// use std::hint::unreachable_unchecked;
///
/// // `b.saturating_add(1)` is always positive (not zero),
/// // hence `checked_div` will never return `None`.
/// // Therefore, the else branch is unreachable.
/// a.checked_div(b.saturating_add(1))
/// .unwrap_or_else(|| unsafe { unreachable_unchecked() })
/// }
///
/// assert_eq!(div_1(7, 0), 7);
/// assert_eq!(div_1(9, 1), 4);
/// assert_eq!(div_1(11, u32::MAX), 0);
/// ```
#[inline]
#[stable(feature = "unreachable", since = "1.27.0")]
#[rustc_const_stable(feature = "const_unreachable_unchecked", since = "1.57.0")]
#[cfg_attr(miri, track_caller)] // even without panics, this helps for Miri backtraces
pub const unsafe fn unreachable_unchecked() -> ! {
ub_checks::assert_unsafe_precondition!(
check_language_ub,
"hint::unreachable_unchecked must never be reached",
() => false
);
// SAFETY: the safety contract for `intrinsics::unreachable` must
// be upheld by the caller.
unsafe { intrinsics::unreachable() }
}
/// Makes a *soundness* promise to the compiler that `cond` holds.
///
/// This may allow the optimizer to simplify things,
/// but it might also make the generated code slower.
/// Either way, calling it will most likely make compilation take longer.
///
/// This is a situational tool for micro-optimization, and is allowed to do nothing.
/// Any use should come with a repeatable benchmark to show the value
/// and allow removing it later should the optimizer get smarter and no longer need it.
///
/// The more complicated the condition the less likely this is to be fruitful.
/// For example, `assert_unchecked(foo.is_sorted())` is a complex enough value
/// that the compiler is unlikely to be able to take advantage of it.
///
/// There's also no need to `assert_unchecked` basic properties of things. For
/// example, the compiler already knows the range of `count_ones`, so there's no
/// benefit to `let n = u32::count_ones(x); assert_unchecked(n <= u32::BITS);`.
///
/// If ever you're tempted to write `assert_unchecked(false)`, then you're
/// actually looking for [`unreachable_unchecked()`].
///
/// You may know this from other places
/// as [`llvm.assume`](https://llvm.org/docs/LangRef.html#llvm-assume-intrinsic)
/// or [`__builtin_assume`](https://clang.llvm.org/docs/LanguageExtensions.html#builtin-assume).
///
/// This promotes a correctness requirement to a soundness requirement.
/// Don't do that without very good reason.
///
/// # Safety
///
/// `cond` must be `true`. It's immediate UB to call this with `false`.
///
#[inline(always)]
#[doc(alias = "assume")]
#[track_caller]
#[unstable(feature = "hint_assert_unchecked", issue = "119131")]
#[rustc_const_unstable(feature = "const_hint_assert_unchecked", issue = "119131")]
pub const unsafe fn assert_unchecked(cond: bool) {
// SAFETY: The caller promised `cond` is true.
unsafe {
ub_checks::assert_unsafe_precondition!(
check_language_ub,
"hint::assert_unchecked must never be called when the condition is false",
(cond: bool = cond) => cond,
);
crate::intrinsics::assume(cond);
}
}
/// Emits a machine instruction to signal the processor that it is running in
/// a busy-wait spin-loop ("spin lock").
///
/// Upon receiving the spin-loop signal the processor can optimize its behavior by,
/// for example, saving power or switching hyper-threads.
///
/// This function is different from [`thread::yield_now`] which directly
/// yields to the system's scheduler, whereas `spin_loop` does not interact
/// with the operating system.
///
/// A common use case for `spin_loop` is implementing bounded optimistic
/// spinning in a CAS loop in synchronization primitives. To avoid problems
/// like priority inversion, it is strongly recommended that the spin loop is
/// terminated after a finite amount of iterations and an appropriate blocking
/// syscall is made.
///
/// **Note**: On platforms that do not support receiving spin-loop hints this
/// function does not do anything at all.
///
/// # Examples
///
/// ```
/// use std::sync::atomic::{AtomicBool, Ordering};
/// use std::sync::Arc;
/// use std::{hint, thread};
///
/// // A shared atomic value that threads will use to coordinate
/// let live = Arc::new(AtomicBool::new(false));
///
/// // In a background thread we'll eventually set the value
/// let bg_work = {
/// let live = live.clone();
/// thread::spawn(move || {
/// // Do some work, then make the value live
/// do_some_work();
/// live.store(true, Ordering::Release);
/// })
/// };
///
/// // Back on our current thread, we wait for the value to be set
/// while !live.load(Ordering::Acquire) {
/// // The spin loop is a hint to the CPU that we're waiting, but probably
/// // not for very long
/// hint::spin_loop();
/// }
///
/// // The value is now set
/// # fn do_some_work() {}
/// do_some_work();
/// bg_work.join()?;
/// # Ok::<(), Box<dyn core::any::Any + Send + 'static>>(())
/// ```
///
/// [`thread::yield_now`]: ../../std/thread/fn.yield_now.html
#[inline(always)]
#[stable(feature = "renamed_spin_loop", since = "1.49.0")]
pub fn spin_loop() {
#[cfg(target_arch = "x86")]
{
// SAFETY: the `cfg` attr ensures that we only execute this on x86 targets.
unsafe { crate::arch::x86::_mm_pause() };
}
#[cfg(target_arch = "x86_64")]
{
// SAFETY: the `cfg` attr ensures that we only execute this on x86_64 targets.
unsafe { crate::arch::x86_64::_mm_pause() };
}
#[cfg(target_arch = "riscv32")]
{
crate::arch::riscv32::pause();
}
#[cfg(target_arch = "riscv64")]
{
crate::arch::riscv64::pause();
}
#[cfg(any(target_arch = "aarch64", target_arch = "arm64ec"))]
{
// SAFETY: the `cfg` attr ensures that we only execute this on aarch64 targets.
unsafe { crate::arch::aarch64::__isb(crate::arch::aarch64::SY) };
}
#[cfg(all(target_arch = "arm", target_feature = "v6"))]
{
// SAFETY: the `cfg` attr ensures that we only execute this on arm targets
// with support for the v6 feature.
unsafe { crate::arch::arm::__yield() };
}
}
/// An identity function that *__hints__* to the compiler to be maximally pessimistic about what
/// `black_box` could do.
///
/// Unlike [`std::convert::identity`], a Rust compiler is encouraged to assume that `black_box` can
/// use `dummy` in any possible valid way that Rust code is allowed to without introducing undefined
/// behavior in the calling code. This property makes `black_box` useful for writing code in which
/// certain optimizations are not desired, such as benchmarks.
///
/// Note however, that `black_box` is only (and can only be) provided on a "best-effort" basis. The
/// extent to which it can block optimisations may vary depending upon the platform and code-gen
/// backend used. Programs cannot rely on `black_box` for *correctness*, beyond it behaving as the
/// identity function. As such, it **must not be relied upon to control critical program behavior.**
/// This _immediately_ precludes any direct use of this function for cryptographic or security
/// purposes.
///
/// [`std::convert::identity`]: crate::convert::identity
///
/// # When is this useful?
///
/// While not suitable in those mission-critical cases, `black_box`'s functionality can generally be
/// relied upon for benchmarking, and should be used there. It will try to ensure that the
/// compiler doesn't optimize away part of the intended test code based on context. For
/// example:
///
/// ```
/// fn contains(haystack: &[&str], needle: &str) -> bool {
/// haystack.iter().any(|x| x == &needle)
/// }
///
/// pub fn benchmark() {
/// let haystack = vec!["abc", "def", "ghi", "jkl", "mno"];
/// let needle = "ghi";
/// for _ in 0..10 {
/// contains(&haystack, needle);
/// }
/// }
/// ```
///
/// The compiler could theoretically make optimizations like the following:
///
/// - The `needle` and `haystack` do not change, move the call to `contains` outside the loop and
/// delete the loop
/// - Inline `contains`
/// - `needle` and `haystack` have values known at compile time, `contains` is always true. Remove
/// the call and replace with `true`
/// - Nothing is done with the result of `contains`: delete this function call entirely
/// - `benchmark` now has no purpose: delete this function
///
/// It is not likely that all of the above happens, but the compiler is definitely able to make some
/// optimizations that could result in a very inaccurate benchmark. This is where `black_box` comes
/// in:
///
/// ```
/// use std::hint::black_box;
///
/// // Same `contains` function
/// fn contains(haystack: &[&str], needle: &str) -> bool {
/// haystack.iter().any(|x| x == &needle)
/// }
///
/// pub fn benchmark() {
/// let haystack = vec!["abc", "def", "ghi", "jkl", "mno"];
/// let needle = "ghi";
/// for _ in 0..10 {
/// // Adjust our benchmark loop contents
/// black_box(contains(black_box(&haystack), black_box(needle)));
/// }
/// }
/// ```
///
/// This essentially tells the compiler to block optimizations across any calls to `black_box`. So,
/// it now:
///
/// - Treats both arguments to `contains` as unpredictable: the body of `contains` can no longer be
/// optimized based on argument values
/// - Treats the call to `contains` and its result as volatile: the body of `benchmark` cannot
/// optimize this away
///
/// This makes our benchmark much more realistic to how the function would actually be used, where
/// arguments are usually not known at compile time and the result is used in some way.
#[inline]
#[stable(feature = "bench_black_box", since = "1.66.0")]
#[rustc_const_unstable(feature = "const_black_box", issue = "none")]
pub const fn black_box<T>(dummy: T) -> T {
crate::intrinsics::black_box(dummy)
}
/// An identity function that causes an `unused_must_use` warning to be
/// triggered if the given value is not used (returned, stored in a variable,
/// etc) by the caller.
///
/// This is primarily intended for use in macro-generated code, in which a
/// [`#[must_use]` attribute][must_use] either on a type or a function would not
/// be convenient.
///
/// [must_use]: https://doc.rust-lang.org/reference/attributes/diagnostics.html#the-must_use-attribute
///
/// # Example
///
/// ```
/// #![feature(hint_must_use)]
///
/// use core::fmt;
///
/// pub struct Error(/* ... */);
///
/// #[macro_export]
/// macro_rules! make_error {
/// ($($args:expr),*) => {
/// core::hint::must_use({
/// let error = $crate::make_error(core::format_args!($($args),*));
/// error
/// })
/// };
/// }
///
/// // Implementation detail of make_error! macro.
/// #[doc(hidden)]
/// pub fn make_error(args: fmt::Arguments<'_>) -> Error {
/// Error(/* ... */)
/// }
///
/// fn demo() -> Option<Error> {
/// if true {
/// // Oops, meant to write `return Some(make_error!("..."));`
/// Some(make_error!("..."));
/// }
/// None
/// }
/// #
/// # // Make rustdoc not wrap the whole snippet in fn main, so that $crate::make_error works
/// # fn main() {}
/// ```
///
/// In the above example, we'd like an `unused_must_use` lint to apply to the
/// value created by `make_error!`. However, neither `#[must_use]` on a struct
/// nor `#[must_use]` on a function is appropriate here, so the macro expands
/// using `core::hint::must_use` instead.
///
/// - We wouldn't want `#[must_use]` on the `struct Error` because that would
/// make the following unproblematic code trigger a warning:
///
/// ```
/// # struct Error;
/// #
/// fn f(arg: &str) -> Result<(), Error>
/// # { Ok(()) }
///
/// #[test]
/// fn t() {
/// // Assert that `f` returns error if passed an empty string.
/// // A value of type `Error` is unused here but that's not a problem.
/// f("").unwrap_err();
/// }
/// ```
///
/// - Using `#[must_use]` on `fn make_error` can't help because the return value
/// *is* used, as the right-hand side of a `let` statement. The `let`
/// statement looks useless but is in fact necessary for ensuring that
/// temporaries within the `format_args` expansion are not kept alive past the
/// creation of the `Error`, as keeping them alive past that point can cause
/// autotrait issues in async code:
///
/// ```
/// # #![feature(hint_must_use)]
/// #
/// # struct Error;
/// #
/// # macro_rules! make_error {
/// # ($($args:expr),*) => {
/// # core::hint::must_use({
/// # // If `let` isn't used, then `f()` produces a non-Send future.
/// # let error = make_error(core::format_args!($($args),*));
/// # error
/// # })
/// # };
/// # }
/// #
/// # fn make_error(args: core::fmt::Arguments<'_>) -> Error {
/// # Error
/// # }
/// #
/// async fn f() {
/// // Using `let` inside the make_error expansion causes temporaries like
/// // `unsync()` to drop at the semicolon of that `let` statement, which
/// // is prior to the await point. They would otherwise stay around until
/// // the semicolon on *this* statement, which is after the await point,
/// // and the enclosing Future would not implement Send.
/// log(make_error!("look: {:p}", unsync())).await;
/// }
///
/// async fn log(error: Error) {/* ... */}
///
/// // Returns something without a Sync impl.
/// fn unsync() -> *const () {
/// 0 as *const ()
/// }
/// #
/// # fn test() {
/// # fn assert_send(_: impl Send) {}
/// # assert_send(f());
/// # }
/// ```
#[unstable(feature = "hint_must_use", issue = "94745")]
#[rustc_const_unstable(feature = "hint_must_use", issue = "94745")]
#[must_use] // <-- :)
#[inline(always)]
pub const fn must_use<T>(value: T) -> T {
value
}