Euclidean distance

# dist.esq
fn dist[N](x: f32[N], y: f32[N]) -> f32 = {
    let d  = x .- y;
    let sq = d .* d;
    +/(sq)         # squared distance
}

fn main() -> i32 = {
    let a = [3.0, 4.0];
    let b = [0.0, 0.0];
    dist(a, b) as i32         # exit 25 (sqrt would be 5)
}

$ ./esquec build dist.esq -o dist
$ ./dist; echo $?
25

Notes

• dist returns the squared distance because there is no sqrt intrinsic today. (You can implement Newton's method with iterate_until if you really want one — see the iterate-until example.)
• The body is a chain of three tensor expressions: subtract, square, reduce. The compiler emits SIMD for each step independently and then fuses where it can.

Pipeline form

The same function written with |>:

fn dist[N](x: f32[N], y: f32[N]) -> f32 = {
    let d = x .- y;
    d .* d |> sumf
}

fn sumf[N](v: f32[N]) -> f32 = +/(v)

For a three-step pipeline this is borderline; +/(d .* d) is shorter. Reach for |> when the steps have meaningful names.