time: use bit manipulation instead of modulo

[samueltardieu]

time: resolve TODO

Motivation

Existing TODO comment in src/time/driver/wheel/level.rs.

Solution

level_range() always return a strictly positive power of 2. If b is a
strictly positive power of 2, a - (a % b) is equal to a & !(b - 1).

[hawkw]

this looks correct to me. it might be worth commenting on the bit manipulation?

[rneswold]

Does this actually improve performance? Or is the performance the same because the compiler already applied your recommendation? Compilers are very smart and could do this optimization for us without having to obfuscating the code.

In the Rust playground:

#![crate_type = "staticlib"]

#[no_mangle]
pub fn f(x: u32) -> u32 {
    x - (x % 11)
}

shows it loading a register with 11 and performing a division. But in this:

#![crate_type = "staticlib"]

#[no_mangle]
pub fn f(x: u32) -> u32 {
    x - (x % 4)
}

the compiler instead masks the value with 3 before subtracting, so it performed the optimization you described.

[Darksonn]

I don't know, but it would require that the compiler can tell that level_range returns a power of two.

[rneswold]

I see.

I'm not a fan of having comments enforce a constraint (line 39: "Being a power of 2 is important.")

How about, at line 40:

const LEVELS: usize = 6;
const LEVEL_MULT: usize = 1 << LEVELS;

[Darksonn]

Feel free to submit a PR that makes the code easier to read.