`AtomicUsize::from_mut` has incorrect alignment requirements in docs

[CatsAreFluffy]

Location

core::sync::atomic::AtomicUsize::from_mut
core::sync::atomic::AtomicIsize::from_mut

Summary

The documentation for AtomicUsize::from_mut claims that it's only available when usize has alignment 8, but in reality it's available when usize has alignment equal to its size (AFAICT), which is different on non-64-bit platforms. For example, this code compiles on 32-bit Linux:

#![feature(atomic_from_mut)]

use std::sync::atomic::AtomicUsize;

pub fn main(){
    let mut x: usize = 0;
    assert_eq!(std::mem::size_of_val(&x), 4);
    let _y: &mut AtomicUsize = AtomicUsize::from_mut(&mut x);
}

Godbolt

[RodBurman]

The documentation defines Usize as "The pointer-sized unsigned integer type." and further says "The size of this primitive is how many bytes it takes to reference any location in memory. For example, on a 32 bit target, this is 4 bytes and on a 64 bit target, this is 8 bytes."
This it is inherently defined by the target architecture given the architecture supports atomic loads & stores. (as documented). It therefore does seem odd that it would be restricted to 8byte/64bit architectures.

[ARandomDev99]

@rustbot claim