// Comment about https://mastodon.social/@brandewinder@hachyderm.io/113064780292323247
// https://brandewinder.com/2024/09/01/should-i-use-simd-vectors/
using System.Numerics;
using System.Runtime.InteropServices;
using System.Runtime.Intrinsics;

public class VectorProcessor
{
    public static float Take2(float[] left, float[] right)
    {
        var size = left.Length;
        var s1 = new ReadOnlySpan<float>(left);
        var s2 = new ReadOnlySpan<float>(right);
        var total = 0.0f;
        // Use Vector<float>.Count, don't hardcode
        for (int i = 0; i < size / Vector<float>.Count; i++)
        {
            var s = Vector<float>.Count * i;
            var v1 = new Vector<float>(s1.Slice(s, Vector<float>.Count));
            var v2 = new Vector<float>(s2.Slice(s, Vector<float>.Count));
            var diff = v1 - v2;
            total += Vector.Dot(diff, diff);
        }

        return MathF.Sqrt(total);
    }

    public static float TakeFaster(float[] left, float[] right)
    {
        var s1 = MemoryMarshal.Cast<float, Vector<float>>(new ReadOnlySpan<float>(left));
        var s2 = MemoryMarshal.Cast<float, Vector<float>>(new ReadOnlySpan<float>(right));
        var total = 0.0f;
        for (int i = 0; i < s1.Length && i < s2.Length; i++)
        {
            var v1 = s1[i];
            var v2 = s2[i];
            var diff = v1 - v2;
            total += Vector.Dot(diff, diff);
        }

        return MathF.Sqrt(total);
    }
}