wrapping_{op} is not inlined in debug mode

[Lokathor]

The wrapping operations for numbers, such as wrapping_mul for u32 or i32, are marked #[inline] instead of #[inline(always)]. This gives them bad performance in debug builds.

#[inline(always)]
pub fn a(x: u32, y: u32) -> u32 {
    x.wrapping_mul(y)
}

pub fn b(x: u32, y: u32) -> u32 {
    a(x.wrapping_add(3), y.wrapping_add(2))
}

Output:

example::b:
        sub     rsp, 24
        mov     eax, 3
        mov     dword ptr [rsp + 20], esi
        mov     esi, eax
        call    core::num::<impl u32>::wrapping_add
        mov     dword ptr [rsp + 16], eax
        mov     edi, dword ptr [rsp + 20]
        mov     esi, 2
        call    core::num::<impl u32>::wrapping_add
        mov     dword ptr [rsp + 12], eax
        mov     edi, dword ptr [rsp + 16]
        mov     esi, dword ptr [rsp + 12]
        call    core::num::<impl u32>::wrapping_mul
        mov     dword ptr [rsp + 8], eax
        mov     eax, dword ptr [rsp + 8]
        add     rsp, 24
        ret

1.50.0: https://rust.godbolt.org/z/b6T1eM
Nightly: https://rust.godbolt.org/z/6oMebd

[nagisa]

Duplicate of #75598