#include <bits/stdc++.h>
#include <cassert>
using namespace std;
using ll = long long int;
using u64 = unsigned long long;
using pll = pair<ll, ll>;
// #include <atcoder/all>
// using namespace atcoder;
#define REP(i, a, b) for (ll i = (a); i < (b); i++)
#define REPrev(i, a, b) for (ll i = (a); i >= (b); i--)
#define ALL(coll) (coll).begin(), (coll).end()
#define SIZE(v) ((ll)((v).size()))
#define REPOUT(i, a, b, exp, sep) REP(i, (a), (b)) cout << (exp) << (i + 1 == (b) ? "" : (sep)); cout << "\n"

// @@ !! LIM(tree mod)

// ---- inserted library file tree.cc

struct function_error : runtime_error {
  function_error(const string& msg) : runtime_error(msg) {}
};

struct Tree {

  struct pe_t {
    ll peer;
    ll edge;
    pe_t(ll peer_ = -1, ll edge_ = -1) : peer(peer_), edge(edge_) {}
    static const pe_t end_object;
  };

  struct nbr_t {
    ll parent_idx;                 // pe[parent_idx] is the parent
    vector<pe_t> pe;
    nbr_t() : parent_idx(-1), pe() {}
  };

  template<bool get_peer, bool excl_parent>
  struct nbr_iterator {
    const nbr_t& body;
    ll pe_idx;
    explicit nbr_iterator(const nbr_t& body_, ll pe_idx_) : body(body_), pe_idx(pe_idx_) {
      if constexpr (excl_parent) {
        if (pe_idx == body.parent_idx) pe_idx++;
      }
    }
    auto operator*() const -> typename conditional<get_peer, ll, const pe_t&>::type {
      if constexpr (get_peer) return body.pe[pe_idx].peer;
      else                    return body.pe[pe_idx];
    }
    const nbr_iterator& operator++() {
      pe_idx++;
      if constexpr (excl_parent) {
        if (pe_idx == body.parent_idx) pe_idx++;
      }
      return *this;
    }
    bool operator !=(const nbr_iterator& o) const { return pe_idx != o.pe_idx; }
  };

  template<bool get_peer, bool excl_parent = true>
  struct children_view {
    const nbr_t& body;
    children_view(const nbr_t& body_) : body(body_) {}
    nbr_iterator<get_peer, excl_parent> begin() const { return nbr_iterator<get_peer, excl_parent>(body, 0); }
    nbr_iterator<get_peer, excl_parent> end() { return nbr_iterator<get_peer, excl_parent>(body, std::ssize(body.pe));  }
  };

  ll numNodes;
  ll root;
  vector<nbr_t> _nbr;
  vector<pair<ll, ll>> _edges;   // (x, y) in _edges => x < y
  vector<ll> _stsize;
  vector<ll> _depth;
  vector<ll> _euler_in;
  vector<ll> _euler_out;
  vector<pair<ll, bool>> _euler_edge;
  vector<vector<vector<ll>>> _lca_tbl;

  constexpr static bool use_depth = true;
  constexpr static bool use_stsize = true;
  constexpr static bool use_euler = true;

  Tree(ll numNodes_, ll root_ = 0) : numNodes(numNodes_), root(root_), _nbr(numNodes_) {
    if (numNodes == 1) _set_parent();
  }

  ll add_edge(ll x, ll y) {
    ll i = ssize(_edges);
    if (i >= numNodes - 1) throw range_error("add_edge");
    _nbr[x].pe.emplace_back(y, i);
    _nbr[y].pe.emplace_back(x, i);
    _edges.emplace_back(min(x, y), max(x, y));
    if (i + 1 == numNodes - 1) _set_parent();
    return i;
  }

  void _set_parent() {   // called from constructor, add_edge() and change_root()

    _nbr[root].parent_idx = ssize(_nbr[root].pe);

    if constexpr (use_depth) _depth.resize(numNodes);
    if constexpr (use_stsize) _stsize.resize(numNodes);
    if constexpr (use_euler) {
      _euler_in.resize(numNodes);
      _euler_out.resize(numNodes);
      _euler_edge.resize(2 * numNodes);
    }
    ll euler_idx = 0;

    auto dfs = [&](auto rF, ll nd, ll pt, ll edge) -> void {
      if constexpr (use_depth) _depth[nd] = pt == -1 ? 0 : _depth[pt] + 1;
      if constexpr (use_stsize) _stsize[nd] = 1;
      if constexpr (use_euler) {
        _euler_edge[euler_idx] = {edge, 0};
        _euler_in[nd] = euler_idx;
        euler_idx++;
      }
      for (ll i = 0; i < ssize(_nbr[nd].pe); i++) {
        auto [c_id, c_eg] = _nbr[nd].pe[i];
        if (c_id == pt) _nbr[nd].parent_idx = i;
        else {
          rF(rF, c_id, nd, c_eg);
          if constexpr (use_stsize) _stsize[nd] += _stsize[c_id];
        }
      }
      if constexpr (use_euler) {
        _euler_edge[euler_idx] = {edge, 1};
        _euler_out[nd] = euler_idx;
        euler_idx++;
      }
    };

    dfs(dfs, root, -1, numNodes - 1);
  }

  pe_t parent_pe(ll nd) { return nd == root ? pe_t(-1, -1) : _nbr[nd].pe[_nbr[nd].parent_idx]; }
  ll parent(ll nd) { return nd == root ? -1 : parent_pe(nd).peer; }
  ll num_children(ll nd) { return _nbr[nd].pe.size() - (_nbr[nd].parent_idx == (ll)_nbr[nd].pe.size() ? 0 : 1); }
  pe_t child_pe(ll nd, ll idx) { return _nbr[nd].pe[idx < _nbr[nd].parent_idx ? idx : idx + 1]; }
  ll child(ll nd, ll idx) { return child_pe(nd, idx).peer; }
  ll child_edge(ll nd, ll idx) { return child_pe(nd, idx).edge; }
  auto children_pe(ll nd) { return children_view<false>(_nbr[nd]); }
  auto children(ll nd) { return children_view<true>(_nbr[nd]); }
  auto neighbor_pe(ll nd) { return children_view<false, false>(_nbr[nd]); }
  auto neighbor(ll nd) { return children_view<true, false>(_nbr[nd]); }

  ll stsize(ll nd) {
    if constexpr (use_stsize) return _stsize[nd];
    else throw function_error("use_stsize should be set to call stsize.");
  }

  ll depth(ll nd) {
    if constexpr (use_depth) return _depth[nd];
    else throw function_error("use_depth should be set to call depth.");
  }

  ll _enc_node_pair(ll x, ll y) { return (x + 1) * (numNodes + 1) + (y + 1); }

  ll edge_idx(ll x, ll y) {
    auto [py, ey] = parent_pe(y);
    if (x == py) return ey;
    auto [px, ex] = parent_pe(x);
    if (y == px) return ex;
    return -1LL;
  }

  pair<ll, ll> nodes_of_edge(ll e, ll mode = 0) {
    if (mode == -1) {
      return _edges[e];
    }else {
      const auto& [x, y] = _edges[e];
      if ((x == parent(y)) == (mode == 0)) return {x, y};
      else return {y, x};
    }
  }

  void _set_euler() {
    _euler_in.resize(numNodes);
    _euler_out.resize(numNodes);
    vector<pair<ll, ll>> stack{{root, -1}};
    ll idx = 0;
    while (not stack.empty()) {
      auto& [nd, cidx] = stack.back();
      if (cidx == -1) _euler_in[nd] = idx++;
      cidx++;
      if (cidx < num_children(nd)) stack.emplace_back(child(nd, cidx), -1);
      else {
        _euler_out[nd] = idx++;
        stack.pop_back();
      }
    }
  }

  ll euler_in(ll nd) {
    if constexpr (use_euler) return _euler_in[nd];
    else throw function_error("use_euler should be set to call euler_in.");
  }

  ll euler_out(ll nd) {
    if constexpr (use_euler) return _euler_out[nd];
    else throw function_error("use_euler should be set to call euler_out.");
  }

  tuple<ll, ll, ll> euler_edge(ll idx) {
    if constexpr (use_euler) {
      if (idx == 0) return {numNodes - 1, -1, root};
      else if (idx == 2 * numNodes - 1) return {numNodes - 1, root, -1};
      else {
        auto [e, b] = _euler_edge[idx];
        auto [x, y] = nodes_of_edge(e, b);
        return {e, x, y};
      }
    }
    else throw function_error("use_euler should be set to call euler_out.");
  }

  // Lowest Common Ancestor
  ll lca(ll x, ll y) {
    ll kmax = 1 + bit_width((unsigned)numNodes);
    ll lastmove = 2 * numNodes - 2;
    if (_lca_tbl.empty()) {
      auto choose = [&](const auto& vec, ll a, ll b) -> ll {
        if (0 <= b and b <= lastmove and vec[b] >= 0) return depth(vec[a]) < depth(vec[b]) ? vec[a] : vec[b];
        else return -1;
      };
      _lca_tbl.resize(kmax + 1, vector(2, vector(lastmove + 1, -1LL)));
      for (ll s = 0; s < 2; s++) for (ll i = 0; i <= lastmove; i++) _lca_tbl[0][s][i] = get<2>(euler_edge(i));
      for (ll k = 1; k <= kmax; k++) {
        ll prev_len = 1 << (k - 1);
        for (ll s = 0; s < 2; s++) {
          for (ll i = 0; i <= lastmove; i++) _lca_tbl[k][0][i] = choose(_lca_tbl[k - 1][0], i, i + prev_len);
          for (ll i = 0; i <= lastmove; i++) _lca_tbl[k][1][i] = choose(_lca_tbl[k - 1][1], i, i - prev_len);
        }
      }
    }
    ll a = euler_in(x), b = euler_in(y);
    if (a > b) swap(a, b);
    ll k = countr_zero(bit_floor((unsigned)(b - a + 1)));
    ll i = _lca_tbl[k][0][a];
    ll j = _lca_tbl[k][1][b];
    return depth(i) < depth(j) ? i : j;
  }

  // the path between two nodes (list of nodes, including x and y)
  vector<ll> nnpath(ll x, ll y) {
    vector<ll> ret;
    ll c = lca(x, y);
    for ( ; x != c; x = parent(x)) ret.push_back(x);
    ret.push_back(c);
    ll len = (ll)ret.size();
    for ( ; y != c; y = parent(y)) ret.push_back(y);
    reverse(ret.begin() + len, ret.end());
    return ret;
  }

#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wunused-but-set-variable"    
  tuple<ll, ll, ll, ll, ll> diameter() {
    if (numNodes == 1) return {0, 0, 0, 0, 0};
    if (numNodes == 2) return {1, 0, 1, 0, 1};
    depth(root);   // to ensure that _depth is correctly built
    ll nd0 = max_element(_depth.begin(), _depth.end()) - _depth.begin();
    ll nd1 = -1, ct0 = -1, ct1 = -1;
    ll diam = 0;
    auto dfs2 = [&](auto rF, ll nd, ll dp, ll pt) -> bool {
      // DFS from nd0, which is different from the root.
      bool ret = false;
      ll numChildren = 0;
      for (auto [cld, _e] : _nbr[nd].pe) {
        if (cld == pt) continue;
        numChildren++;
        bool bbb = rF(rF, cld, dp + 1, nd);
        ret = ret || bbb;
      }
      if (numChildren > 0) {
        if (ret) {
          if (diam % 2 == 0) {
            if (dp == diam / 2) ct0 = ct1 = nd;
          }else {
            if (dp == diam / 2) ct0 = nd;
            else if (dp == diam / 2 + 1) ct1 = nd;
          }
        }
      }else {
        if (dp > diam) {
          diam = dp;
          nd1 = nd;
          ret = true;
        }
      }
      return ret;
    };
    dfs2(dfs2, nd0, 0, -1);
    return {diam, nd0, nd1, ct0, ct1};
  }
#pragma GCC diagnostic pop

  pair<ll, ll> centroids() {
    auto dfs = [&](auto rF, ll nd) -> pair<ll, ll> {
      for (ll c : children(nd)) {
        ll a = 2 * stsize(c);
        if (a > numNodes) return rF(rF, c);
        if (a == numNodes) return make_pair(nd, c);
      }
      return make_pair(nd, -1);
    };
    return dfs(dfs, root);
  }

  void change_root(ll newRoot) {
    _stsize.clear();
    _depth.clear();
    _euler_in.clear();
    _euler_out.clear();
    _lca_tbl.clear();

    root = newRoot;
    _set_parent();
  }

};

const Tree::pe_t end_object{-1, -1};

template <typename M>
auto reroot(Tree& tree, const M& unit, auto add, auto mod1, auto mod2) {
  using A = decltype(mod2(M(), 0));
  vector<A> result(tree.numNodes);
  vector<vector<M>> sum_left(tree.numNodes);
  vector<vector<M>> sum_right(tree.numNodes);
  
  auto dfs1 = [&](const auto& recF, ll nd) -> A {
    ll k = tree.num_children(nd);
    vector<M> ws(k);
    for (ll i = 0; i < k; i++) {
      ll c = tree.child(nd, i);
      ws[i] = mod1(recF(recF, c), nd, c);
    }
    sum_left[nd].resize(k + 1, unit);
    sum_right[nd].resize(k + 1, unit);
    for (ll i = 0; i < k; i++) sum_left[nd][i + 1] = add(sum_left[nd][i], ws[i]);
    for (ll i = k - 1; i >= 0; i--) sum_right[nd][i] = add(sum_right[nd][i + 1], ws[i]);
    return mod2(sum_right[nd][0], nd);
  };
  dfs1(dfs1, tree.root);

  auto dfs2 = [&](const auto& recF, ll nd, const M& t) -> void {
    result[nd] = mod2(add(sum_right[nd][0], t), nd);
    ll k = tree.num_children(nd);
    for (ll i = 0; i < k; i++) {
      ll c = tree.child(nd, i);
      M excl_c = add(sum_left[nd][i], sum_right[nd][i + 1]);
      M m_for_c = add(excl_c, t);
      A v_for_c = mod2(m_for_c, nd);
      M pass_c = mod1(v_for_c, c, nd);
      recF(recF, c, pass_c);
    }
  };
  dfs2(dfs2, tree.root, unit);
  
  return result;
}

template <typename M>
vector<M> reroot(Tree& tree, const M& unit, auto add, auto mod1) {
  return reroot<M>(tree, unit, add, mod1, [](const M& m, ll i) -> M { return m; });
}

// ---- end tree.cc

// ---- inserted library file algOp.cc

// Common definitions
//    zero, one, inverse

template<typename T>
const T zero(const T& t) {
  if constexpr (is_integral_v<T> || is_floating_point_v<T>) { return (T)0; }
  else { return t.zero(); }
}

template<typename T>
const T one(const T& t) {
  if constexpr (is_integral_v<T> || is_floating_point_v<T>) { return (T)1; }
  else { return t.one(); }
}

template<typename T>
const T inverse(const T& t) {
  if constexpr (is_floating_point_v<T>) { return 1.0 / t; }
  else { return t.inverse(); }
}

#ifdef BOOST_MP_CPP_INT_HPP
template<> const cpp_int zero(const cpp_int& t) { return cpp_int(0); }
template<> const cpp_int one(const cpp_int& t) { return cpp_int(1); }
#endif // BOOST_MP_CPP_INT_HPP

// begin -- detection ideom
//    cf. https://blog.tartanllama.xyz/detection-idiom/

namespace tartan_detail {
  template <template <class...> class Trait, class Enabler, class... Args>
  struct is_detected : false_type{};

  template <template <class...> class Trait, class... Args>
  struct is_detected<Trait, void_t<Trait<Args...>>, Args...> : true_type{};
}

template <template <class...> class Trait, class... Args>
using is_detected = typename tartan_detail::is_detected<Trait, void, Args...>::type;

// end -- detection ideom


template<typename T>
// using subst_add_t = decltype(T::subst_add(declval<typename T::value_type &>(), declval<typename T::value_type>()));
using subst_add_t = decltype(T::subst_add);
template<typename T>
using has_subst_add = is_detected<subst_add_t, T>;

template<typename T>
using add_t = decltype(T::add);
template<typename T>
using has_add = is_detected<add_t, T>;

template<typename T>
using subst_mult_t = decltype(T::subst_mult);
template<typename T>
using has_subst_mult = is_detected<subst_mult_t, T>;

template<typename T>
using mult_t = decltype(T::mult);
template<typename T>
using has_mult = is_detected<mult_t, T>;

template<typename T>
using subst_subt_t = decltype(T::subst_subt);
template<typename T>
using has_subst_subt = is_detected<subst_subt_t, T>;

template<typename T>
using subt_t = decltype(T::subt);
template<typename T>
using has_subt = is_detected<subt_t, T>;

template <typename Opdef>
struct MyAlg {
  using T = typename Opdef::value_type;
  using value_type = T;
  T v;
  MyAlg() {}
  MyAlg(const T& v_) : v(v_) {}
  MyAlg(T&& v_) : v(move(v_)) {}
  bool operator==(MyAlg o) const { return v == o.v; }
  bool operator!=(MyAlg o) const { return v != o.v; }
  operator T() const { return v; }
  MyAlg zero() const { return MyAlg(Opdef::zero(v)); }
  MyAlg one() const { return MyAlg(Opdef::one(v)); }
  MyAlg inverse() const { return MyAlg(Opdef::inverse(v)); }
  MyAlg operator/=(const MyAlg& o) { return *this *= o.inverse(); }
  MyAlg operator/(const MyAlg& o) const { return (*this) * o.inverse(); }
  MyAlg operator-() const { return zero() - *this; }

  MyAlg& operator +=(const MyAlg& o) { 
    if constexpr (has_subst_add<Opdef>::value) {
      Opdef::subst_add(v, o.v);
      return *this;
    }else if constexpr (has_add<Opdef>::value) {
      v = Opdef::add(v, o.v);
      return *this;
    }else static_assert("either subst_add or add is needed.");

  }
  MyAlg operator +(const MyAlg& o) const { 
    if constexpr (has_add<Opdef>::value) {
      return MyAlg(Opdef::add(v, o.v));
    }else if constexpr (has_subst_add<Opdef>::value) {
      MyAlg ret(v);
      Opdef::subst_add(ret.v, o.v);
      return ret;
    }else static_assert("either subst_add or add is needed.");
  }
  MyAlg& operator *=(const MyAlg& o) { 
    if constexpr (has_subst_mult<Opdef>::value) {
      Opdef::subst_mult(v, o.v);
      return *this;
    }else if constexpr (has_mult<Opdef>::value) {
      v = Opdef::mult(v, o.v);
      return *this;
    }else static_assert("either subst_mult or mult is needed.");

  }
  MyAlg operator *(const MyAlg& o) const { 
    if constexpr (has_mult<Opdef>::value) {
      return MyAlg(Opdef::mult(v, o.v));
    }else if constexpr (has_subst_mult<Opdef>::value) {
      MyAlg ret(v);
      Opdef::subst_mult(ret.v, o.v);
      return ret;
    }else static_assert("either subst_mult or mult is needed.");
  }
  MyAlg& operator -=(const MyAlg& o) { 
    if constexpr (has_subst_subt<Opdef>::value) {
      Opdef::subst_subt(v, o.v);
      return *this;
    }else if constexpr (has_subt<Opdef>::value) {
      v = Opdef::subt(v, o.v);
      return *this;
    }else static_assert("either subst_subt or subt is needed.");

  }
  MyAlg operator -(const MyAlg& o) const { 
    if constexpr (has_subt<Opdef>::value) {
      return MyAlg(Opdef::subt(v, o.v));
    }else if constexpr (has_subst_subt<Opdef>::value) {
      MyAlg ret(v);
      Opdef::subst_subt(ret.v, o.v);
      return ret;
    }else static_assert("either subst_subt or subt is needed.");
  }
  friend istream& operator >>(istream& is, MyAlg& t)       { is >> t.v; return is; }
  friend ostream& operator <<(ostream& os, const MyAlg& t) { os << t.v; return os; }
};





// ---- end algOp.cc

// ---- inserted function f:gcd from util.cc

// auto [g, s, t] = eGCD(a, b)
//     g == gcd(|a|, |b|) and as + bt == g           
//     It guarantees that max(|s|, |t|) <= max(|a| / g, |b| / g)   (when g != 0)
//     Note that gcd(a, 0) == gcd(0, a) == a.
template<typename INT=ll>
tuple<INT, INT, INT> eGCD(INT a, INT b) {
  INT sa = a < 0 ? -1 : 1;
  INT ta = 0;
  INT za = a * sa;
  INT sb = 0;
  INT tb = b < 0 ? -1 : 1;
  INT zb = b * tb;
  while (zb != 0) {
    INT q = za / zb;
    INT r = za % zb;
    za = zb;
    zb = r;
    INT new_sb = sa - q * sb;
    sa = sb;
    sb = new_sb;
    INT new_tb = ta - q * tb;
    ta = tb;
    tb = new_tb;
  }
  return {za, sa, ta};
}

pair<ll, ll> crt_sub(ll a1, ll x1, ll a2, ll x2) {
  // DLOGKL("crt_sub", a1, x1, a2, x2);
  a1 = a1 % x1;
  a2 = a2 % x2;
  auto [g, s, t] = eGCD(x1, -x2);
  ll gq = (a2 - a1) / g;
  ll gr = (a2 - a1) % g;
  if (gr != 0) return {-1, -1};
  s *= gq;
  t *= gq;
  ll z = x1 / g * x2;
  // DLOGK(z);
  s = s % (x2 / g);
  ll r = (x1 * s + a1) % z;
  // DLOGK(r);
  if (r < 0) r += z;
  // DLOGK(r);
  return {r, z};
};

// Chinese Remainder Theorem
//
//    r = crt(a1, x1, a2, x2)
//    ==>   r = a1 (mod x1);  r = a2 (mod x2);  0 <= r < lcm(x1, x2)
//    If no such r exists, returns -1
//    Note: x1 and x2 should >= 1.  a1 and a2 can be negative or zero.
//
//    r = crt(as, xs)
//    ==>   for all i. r = as[i] (mod xs[i]); 0 <= r < lcm(xs)
//    If no such r exists, returns -1
//    Note: xs[i] should >= 1.  as[i] can be negative or zero.
//          It should hold: len(xs) == len(as) > 0

ll crt(ll a1, ll x1, ll a2, ll x2) { return crt_sub(a1, x1, a2, x2).first; }

ll crt(vector<ll> as, vector<ll> xs) {
  // DLOGKL("crt", as, xs);
  assert(xs.size() == as.size() && xs.size() > 0);
  ll r = as[0];
  ll z = xs[0];
  for (size_t i = 1; i < xs.size(); i++) {
    // DLOGK(i, r, z, as[i], xs[i]);
    tie(r, z) = crt_sub(r, z, as[i], xs[i]);
    // DLOGK(r, z);
    if (r == -1) return -1;
  }
  return r;
}

// ---- end f:gcd

// ---- inserted library file mod.cc

template<int mod=0, typename INT=ll>
struct FpG {   // G for General
  static INT dyn_mod;

  static INT getMod() {
    if (mod == 0) return dyn_mod;
    else          return (INT)mod;
  }
  
  // Effective only when mod == 0.
  // _mod must be less than the half of the maximum value of INT.
  static void setMod(INT _mod) {  
    dyn_mod = _mod;
  }

  static INT _conv(INT x) {
    if (x >= getMod())  return x % getMod();
    if (x >= 0)         return x;
    if (x >= -getMod()) return x + getMod();
    INT y = x % getMod();
    if (y == 0) return 0;
    return y + getMod();
  }

  INT val;

  FpG(INT t = 0) : val(_conv(t)) {}
  FpG(const FpG& t) : val(t.val) {}
  FpG& operator =(const FpG& t) { val = t.val; return *this; }
  FpG& operator =(INT t) { val = _conv(t); return *this; }

  FpG& operator +=(const FpG& t) {
    val += t.val;
    if (val >= getMod()) val -= getMod();
    return *this;
  }

  FpG& operator -=(const FpG& t) {
    val -= t.val;
    if (val < 0) val += getMod();
    return *this;
  }

  FpG& operator *=(const FpG& t) {
    val = (val * t.val) % getMod();
    return *this;
  }

  FpG inv() const {
    if (val == 0) { throw runtime_error("FpG::inv(): called for zero."); }
    auto [g, u, v] = eGCD(val, getMod());
    if (g != 1) { throw runtime_error("FpG::inv(): not co-prime."); }
    return FpG(u);
  }

  FpG zero() const { return (FpG)0; }
  FpG one() const { return (FpG)1; }
  FpG inverse() const { return inv(); }

  FpG& operator /=(const FpG& t) {
    return (*this) *= t.inv();
  }

  FpG operator +(const FpG& t) const { return FpG(val) += t; }
  FpG operator -(const FpG& t) const { return FpG(val) -= t; }
  FpG operator *(const FpG& t) const { return FpG(val) *= t; }
  FpG operator /(const FpG& t) const { return FpG(val) /= t; }
  FpG operator -() const { return FpG(-val); }

  bool operator ==(const FpG& t) const { return val == t.val; }
  bool operator !=(const FpG& t) const { return val != t.val; }
  
  operator INT() const { return val; }

  friend FpG operator +(INT x, const FpG& y) { return FpG(x) + y; }
  friend FpG operator -(INT x, const FpG& y) { return FpG(x) - y; }
  friend FpG operator *(INT x, const FpG& y) { return FpG(x) * y; }
  friend FpG operator /(INT x, const FpG& y) { return FpG(x) / y; }
  friend bool operator ==(INT x, const FpG& y) { return FpG(x) == y; }
  friend bool operator !=(INT x, const FpG& y) { return FpG(x) != y; }
  friend FpG operator +(const FpG& x, INT y) { return x + FpG(y); }
  friend FpG operator -(const FpG& x, INT y) { return x - FpG(y); }
  friend FpG operator *(const FpG& x, INT y) { return x * FpG(y); }
  friend FpG operator /(const FpG& x, INT y) { return x / FpG(y); }
  friend bool operator ==(const FpG& x, INT y) { return x == FpG(y); }
  friend bool operator !=(const FpG& x, INT y) { return x != FpG(y); }

  /* The following are needed to avoid warnings in cases such as FpG x; x = 5 + x; rather than x = FpG(5) + x; */
  friend FpG operator +(int x, const FpG& y) { return FpG(x) + y; }
  friend FpG operator -(int x, const FpG& y) { return FpG(x) - y; }
  friend FpG operator *(int x, const FpG& y) { return FpG(x) * y; }
  friend FpG operator /(int x, const FpG& y) { return FpG(x) / y; }
  friend bool operator ==(int x, const FpG& y) { return FpG(x) == y; }
  friend bool operator !=(int x, const FpG& y) { return FpG(x) != y; }
  friend FpG operator +(const FpG& x, int y) { return x + FpG(y); }
  friend FpG operator -(const FpG& x, int y) { return x - FpG(y); }
  friend FpG operator *(const FpG& x, int y) { return x * FpG(y); }
  friend FpG operator /(const FpG& x, int y) { return x / FpG(y); }
  friend bool operator ==(const FpG& x, int y) { return x == FpG(y); }
  friend bool operator !=(const FpG& x, int y) { return x != FpG(y); }

  friend istream& operator>> (istream& is, FpG& t) {
    INT x; is >> x;
    t = x;
    return is;
  }

  friend ostream& operator<< (ostream& os, const FpG& t) {
    os << t.val;
    return os;
  }

};
template<int mod, typename INT>
INT FpG<mod, INT>::dyn_mod;

template<typename T>
class Comb {
  int nMax;
  vector<T> vFact;
  vector<T> vInvFact;
public:
  Comb(int nm) : nMax(nm), vFact(nm+1), vInvFact(nm+1) {
    vFact[0] = 1;
    for (int i = 1; i <= nMax; i++) vFact[i] = i * vFact[i-1];
    vInvFact.at(nMax) = (T)1 / vFact[nMax];
    for (int i = nMax; i >= 1; i--) vInvFact[i-1] = i * vInvFact[i];
  }
  T fact(int n) { return vFact[n]; }
  T binom(int n, int r) {
    if (r < 0 || r > n) return (T)0;
    return vFact[n] * vInvFact[r] * vInvFact[n-r];
  }
  T binom_dup(int n, int r) { return binom(n + r - 1, r); }
  // The number of permutation extracting r from n.
  T perm(int n, int r) {
    return vFact[n] * vInvFact[n-r];
  }
};

constexpr int primeA = 1'000'000'007;
constexpr int primeB = 998'244'353;          // '
using FpA = FpG<primeA, ll>;
using FpB = FpG<primeB, ll>;

// ---- end mod.cc

// @@ !! LIM -- end mark --

using Fp = FpB;

int main(/* int argc, char *argv[] */) {
  ios_base::sync_with_stdio(false);
  cin.tie(nullptr);
  cout << setprecision(20);

  auto solve = [&]() -> Fp {
    ll N; cin >> N;
    Tree tr(N);
    REP(i, 0, N - 1) {
      ll u, v; cin >> u >> v; u--; v--;
      tr.add_edge(u, v);
    }
    vector vec(3, vector(N, 0LL));
    REP(i, 0, N) vec[0][i] = tr.num_children(i);
    REP(k, 1, 3) {
      REP(i, 0, N) {
        for (ll c : tr.children(i)) vec[k][i] += vec[k - 1][c];
      }
    }
    Fp ans = 0;
    REP(i, 0, N) {
      ans += vec[2][i];
      for (ll c : tr.children(i)) {
        ans += Fp(tr.num_children(c)) * (tr.num_children(i) - 1);
      }
    }
    return ans;
  };

  ll T; cin >> T;
  REP(t, 0, T) cout << solve() << "\n";

  return 0;
}