struct MatrixBatch {
  size: vec4<u32>,
  data: array<vec4<f32>>,
}

@group(0) @binding(0)
var<storage, read> m: MatrixBatch;
@group(0) @binding(1)
var<storage, read_write> centered: MatrixBatch;

@compute @workgroup_size(1)
fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
  centered.size = m.size;

  let vc =  m.size.x / 4;
  let ro = (global_id.z * m.size.y + global_id.y) * vc;

  var s = vec4(0f);
  for (var x = 0u; x < vc; x++) {
    s += m.data[ro + x];
  }

  centered.data[ro + global_id.x] = m.data[ro + global_id.x] - vec4(s.x + s.y + s.z + s.w) / f32(m.size.x);
}