Get unchecked

Bounds checks on slice element accesses are not usually a performance problem in programs. But these checks can slow down programs slightly.

In Rust, a direct access of a slice may have bounds checks that are not optimized out. A call to get_unchecked, though, never has bounds checks.

Example benchmark

Consider this Rust program. It creates an vector of 10000 elements, and then assigns a value somewhere in the middle of the vector.

Version 1 We access many elements in the vector's memory and test for the value 5 that was set in the vector.

Version 2 We use get_unchecked() in an unsafe block of code to test elements in the vector. No bounds checks are used in this loop.

Result The performance advantage of using get_unchecked is clear and reproducible. The bounds checks cause a significant performance reduction.

fn main() {
    // Vector for testing.
    let mut x = vec![];
    for _ in 0..10000 {
        x.push(0);
    }
    x[5555] = 5;

    // Version 1: use checked slice accesses.
    let t = Instant::now();
    let mut count = 0;
    for _ in 0..100000 {
        for j in 10..9900 {
            if x[j] == 5 {
                count += 1;
            }
        }
    }
    println!("{}", t.elapsed().as_millis());

    // Version 2: use get_unchecked.
    let t = Instant::now();
    for _ in 0..100000 {
        for j in 10..9900 {
            unsafe {
                if *x.get_unchecked(j) == 5 {
                    count += 1;
                }
            }
        }
    }
    println!("{}", t.elapsed().as_millis());
    println!("{}", count);
}468 ms    x[j]
314 ms    *x.get_unchecked(j)
200000

Some discussion

Calling get_unchecked in Rust program always will involve some risk. It should only be used when you are absolutely certain it is correct.

Info You can call get_unchecked on invalid memory addresses, and unexpected values will be returned. The program may not terminate.

Using get_checked in Rust can significantly speed up some functions that repeatedly read elements from a slice. But the optimization is not safe, and must be carefully considered.