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

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

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

@group(0) @binding(0)
var<storage, read> ps: MatrixBatch;
@group(0) @binding(1)
var<storage, read_write> result: UVectorBatch;

@compute @workgroup_size(1)
fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
  result.size.x = ps.size.y;
  result.size.z = ps.size.z;

  var mi = 0u;
  var mp = 0f;
  var pi = global_id.z * ps.size.y * ps.size.x + global_id.x * ps.size.x;
  for (var i = 0u; i < ps.size.x; i++) {
    if (ps.data[pi] > mp) {
      mi = i;
      mp = ps.data[pi];
    }
    pi++;
  }
  result.data[global_id.z * result.size.x + global_id.x] = mi;
}