Dot product

# dot.esq
fn dot[N](x: f32[N], y: f32[N]) -> f32 = +/(x .* y)

fn main() -> i32 = {
    let a = [1.0, 2.0, 3.0];
    let b = [4.0, 5.0, 6.0];
    let result = dot(a, b);     # 32.0
    result as i32                # exit 32
}

$ ./esquec build dot.esq -o dot
$ ./dot; echo $?
32

Notes

• [N] is a shape parameter. dot is generic over the length of its operands; both arguments must have the same shape.
• .* is element-wise multiplication: [1*4, 2*5, 3*6] = [4,10,18].
• +/ reduces the resulting tensor by addition: 4+10+18 = 32.
• The cast result as i32 truncates toward zero; we use it because exit codes are integers.

Monomorphisation

The call dot(a, b) where both operands have shape [3] causes the compiler to emit a specialised copy dot__3. Add a second use:

fn use4(p: f32[4], q: f32[4]) -> f32 = dot(p, q)

…and the compiler also emits dot__4. dot__3 and dot__4 are independently scheduled and (for SIMD-aligned shapes) use different backends — dot__4 and dot__8 are full SSE/AVX2 paths; dot__3 takes the scalar tail.

You can see all of this with --emit=ceir:

$ ./esquec build dot.esq --emit=ceir | head

Look for the function name dot__3 (it will not say plain dot).

Generalising

A shape-generic rms (root mean square) is one extra line:

fn rms[N](x: f32[N]) -> f32 = {
    let sq = x .* x;
    +/(sq) / 3.0          # hardcoded N=3 until shape-as-value lands
}

Until shape values can flow as runtime scalars (planned), pass N as a separate f32 parameter or use a fixed shape.