#line 1 "/home/maspy/compro/library/my_template.hpp"
#if defined(LOCAL)
#include <my_template_compiled.hpp>
#else

// https://codeforces.com/blog/entry/96344
#pragma GCC optimize("Ofast,unroll-loops")
// いまの CF だとこれ入れると動かない？
// #pragma GCC target("avx2,popcnt")

#include <bits/stdc++.h>

using namespace std;

using ll = long long;
using u8 = uint8_t;
using u16 = uint16_t;
using u32 = uint32_t;
using u64 = uint64_t;
using i128 = __int128;
using u128 = unsigned __int128;
using f128 = __float128;

template <class T>
constexpr T infty = 0;
template <>
constexpr int infty<int> = 1'010'000'000;
template <>
constexpr ll infty<ll> = 2'020'000'000'000'000'000;
template <>
constexpr u32 infty<u32> = infty<int>;
template <>
constexpr u64 infty<u64> = infty<ll>;
template <>
constexpr i128 infty<i128> = i128(infty<ll>) * 2'000'000'000'000'000'000;
template <>
constexpr double infty<double> = infty<ll>;
template <>
constexpr long double infty<long double> = infty<ll>;

using pi = pair<ll, ll>;
using vi = vector<ll>;
template <class T>
using vc = vector<T>;
template <class T>
using vvc = vector<vc<T>>;
template <class T>
using vvvc = vector<vvc<T>>;
template <class T>
using vvvvc = vector<vvvc<T>>;
template <class T>
using vvvvvc = vector<vvvvc<T>>;
template <class T>
using pq = priority_queue<T>;
template <class T>
using pqg = priority_queue<T, vector<T>, greater<T>>;

#define vv(type, name, h, ...) vector<vector<type>> name(h, vector<type>(__VA_ARGS__))
#define vvv(type, name, h, w, ...) vector<vector<vector<type>>> name(h, vector<vector<type>>(w, vector<type>(__VA_ARGS__)))
#define vvvv(type, name, a, b, c, ...) \
  vector<vector<vector<vector<type>>>> name(a, vector<vector<vector<type>>>(b, vector<vector<type>>(c, vector<type>(__VA_ARGS__))))

// https://trap.jp/post/1224/
#define FOR1(a) for (ll _ = 0; _ < ll(a); ++_)
#define FOR2(i, a) for (ll i = 0; i < ll(a); ++i)
#define FOR3(i, a, b) for (ll i = a; i < ll(b); ++i)
#define FOR4(i, a, b, c) for (ll i = a; i < ll(b); i += (c))
#define FOR1_R(a) for (ll i = (a)-1; i >= ll(0); --i)
#define FOR2_R(i, a) for (ll i = (a)-1; i >= ll(0); --i)
#define FOR3_R(i, a, b) for (ll i = (b)-1; i >= ll(a); --i)
#define overload4(a, b, c, d, e, ...) e
#define overload3(a, b, c, d, ...) d
#define FOR(...) overload4(__VA_ARGS__, FOR4, FOR3, FOR2, FOR1)(__VA_ARGS__)
#define FOR_R(...) overload3(__VA_ARGS__, FOR3_R, FOR2_R, FOR1_R)(__VA_ARGS__)

#define FOR_subset(t, s) for (ll t = (s); t >= 0; t = (t == 0 ? -1 : (t - 1) & (s)))
#define all(x) x.begin(), x.end()
#define len(x) ll(x.size())
#define elif else if

#define eb emplace_back
#define mp make_pair
#define mt make_tuple
#define fi first
#define se second

#define stoi stoll

int popcnt(int x) { return __builtin_popcount(x); }
int popcnt(u32 x) { return __builtin_popcount(x); }
int popcnt(ll x) { return __builtin_popcountll(x); }
int popcnt(u64 x) { return __builtin_popcountll(x); }
int popcnt_mod_2(int x) { return __builtin_parity(x); }
int popcnt_mod_2(u32 x) { return __builtin_parity(x); }
int popcnt_mod_2(ll x) { return __builtin_parityll(x); }
int popcnt_mod_2(u64 x) { return __builtin_parityll(x); }
// (0, 1, 2, 3, 4) -> (-1, 0, 1, 1, 2)
int topbit(int x) { return (x == 0 ? -1 : 31 - __builtin_clz(x)); }
int topbit(u32 x) { return (x == 0 ? -1 : 31 - __builtin_clz(x)); }
int topbit(ll x) { return (x == 0 ? -1 : 63 - __builtin_clzll(x)); }
int topbit(u64 x) { return (x == 0 ? -1 : 63 - __builtin_clzll(x)); }
// (0, 1, 2, 3, 4) -> (-1, 0, 1, 0, 2)
int lowbit(int x) { return (x == 0 ? -1 : __builtin_ctz(x)); }
int lowbit(u32 x) { return (x == 0 ? -1 : __builtin_ctz(x)); }
int lowbit(ll x) { return (x == 0 ? -1 : __builtin_ctzll(x)); }
int lowbit(u64 x) { return (x == 0 ? -1 : __builtin_ctzll(x)); }

template <typename T>
T floor(T a, T b) {
  return a / b - (a % b && (a ^ b) < 0);
}
template <typename T>
T ceil(T x, T y) {
  return floor(x + y - 1, y);
}
template <typename T>
T bmod(T x, T y) {
  return x - y * floor(x, y);
}
template <typename T>
pair<T, T> divmod(T x, T y) {
  T q = floor(x, y);
  return {q, x - q * y};
}

template <typename T, typename U>
T SUM(const vector<U> &A) {
  T sm = 0;
  for (auto &&a: A) sm += a;
  return sm;
}

#define MIN(v) *min_element(all(v))
#define MAX(v) *max_element(all(v))
#define LB(c, x) distance((c).begin(), lower_bound(all(c), (x)))
#define UB(c, x) distance((c).begin(), upper_bound(all(c), (x)))
#define UNIQUE(x) sort(all(x)), x.erase(unique(all(x)), x.end()), x.shrink_to_fit()

template <typename T>
T POP(deque<T> &que) {
  T a = que.front();
  que.pop_front();
  return a;
}
template <typename T>
T POP(pq<T> &que) {
  T a = que.top();
  que.pop();
  return a;
}
template <typename T>
T POP(pqg<T> &que) {
  T a = que.top();
  que.pop();
  return a;
}
template <typename T>
T POP(vc<T> &que) {
  T a = que.back();
  que.pop_back();
  return a;
}

template <typename F>
ll binary_search(F check, ll ok, ll ng, bool check_ok = true) {
  if (check_ok) assert(check(ok));
  while (abs(ok - ng) > 1) {
    auto x = (ng + ok) / 2;
    (check(x) ? ok : ng) = x;
  }
  return ok;
}
template <typename F>
double binary_search_real(F check, double ok, double ng, int iter = 100) {
  FOR(iter) {
    double x = (ok + ng) / 2;
    (check(x) ? ok : ng) = x;
  }
  return (ok + ng) / 2;
}

template <class T, class S>
inline bool chmax(T &a, const S &b) {
  return (a < b ? a = b, 1 : 0);
}
template <class T, class S>
inline bool chmin(T &a, const S &b) {
  return (a > b ? a = b, 1 : 0);
}

// ? は -1
vc<int> s_to_vi(const string &S, char first_char) {
  vc<int> A(S.size());
  FOR(i, S.size()) { A[i] = (S[i] != '?' ? S[i] - first_char : -1); }
  return A;
}

template <typename T, typename U>
vector<T> cumsum(vector<U> &A, int off = 1) {
  int N = A.size();
  vector<T> B(N + 1);
  FOR(i, N) { B[i + 1] = B[i] + A[i]; }
  if (off == 0) B.erase(B.begin());
  return B;
}

// stable sort
template <typename T>
vector<int> argsort(const vector<T> &A) {
  vector<int> ids(len(A));
  iota(all(ids), 0);
  sort(all(ids), [&](int i, int j) { return (A[i] == A[j] ? i < j : A[i] < A[j]); });
  return ids;
}

// A[I[0]], A[I[1]], ...
template <typename T>
vc<T> rearrange(const vc<T> &A, const vc<int> &I) {
  vc<T> B(len(I));
  FOR(i, len(I)) B[i] = A[I[i]];
  return B;
}

template <typename T, typename... Vectors>
void concat(vc<T> &first, const Vectors &... others) {
  vc<T> &res = first;
  (res.insert(res.end(), others.begin(), others.end()), ...);
}
#endif
#line 1 "/home/maspy/compro/library/other/io.hpp"
#define FASTIO
#include <unistd.h>

// https://judge.yosupo.jp/submission/21623
namespace fastio {
static constexpr uint32_t SZ = 1 << 17;
char ibuf[SZ];
char obuf[SZ];
char out[100];
// pointer of ibuf, obuf
uint32_t pil = 0, pir = 0, por = 0;

struct Pre {
  char num[10000][4];
  constexpr Pre() : num() {
    for (int i = 0; i < 10000; i++) {
      int n = i;
      for (int j = 3; j >= 0; j--) {
        num[i][j] = n % 10 | '0';
        n /= 10;
      }
    }
  }
} constexpr pre;

inline void load() {
  memcpy(ibuf, ibuf + pil, pir - pil);
  pir = pir - pil + fread(ibuf + pir - pil, 1, SZ - pir + pil, stdin);
  pil = 0;
  if (pir < SZ) ibuf[pir++] = '\n';
}

inline void flush() {
  fwrite(obuf, 1, por, stdout);
  por = 0;
}

void rd(char &c) {
  do {
    if (pil + 1 > pir) load();
    c = ibuf[pil++];
  } while (isspace(c));
}

void rd(string &x) {
  x.clear();
  char c;
  do {
    if (pil + 1 > pir) load();
    c = ibuf[pil++];
  } while (isspace(c));
  do {
    x += c;
    if (pil == pir) load();
    c = ibuf[pil++];
  } while (!isspace(c));
}

template <typename T>
void rd_real(T &x) {
  string s;
  rd(s);
  x = stod(s);
}

template <typename T>
void rd_integer(T &x) {
  if (pil + 100 > pir) load();
  char c;
  do
    c = ibuf[pil++];
  while (c < '-');
  bool minus = 0;
  if constexpr (is_signed<T>::value || is_same_v<T, i128>) {
    if (c == '-') { minus = 1, c = ibuf[pil++]; }
  }
  x = 0;
  while ('0' <= c) { x = x * 10 + (c & 15), c = ibuf[pil++]; }
  if constexpr (is_signed<T>::value || is_same_v<T, i128>) {
    if (minus) x = -x;
  }
}

void rd(int &x) { rd_integer(x); }
void rd(ll &x) { rd_integer(x); }
void rd(i128 &x) { rd_integer(x); }
void rd(u32 &x) { rd_integer(x); }
void rd(u64 &x) { rd_integer(x); }
void rd(u128 &x) { rd_integer(x); }
void rd(double &x) { rd_real(x); }
void rd(long double &x) { rd_real(x); }
void rd(f128 &x) { rd_real(x); }

template <class T, class U>
void rd(pair<T, U> &p) {
  return rd(p.first), rd(p.second);
}
template <size_t N = 0, typename T>
void rd_tuple(T &t) {
  if constexpr (N < std::tuple_size<T>::value) {
    auto &x = std::get<N>(t);
    rd(x);
    rd_tuple<N + 1>(t);
  }
}
template <class... T>
void rd(tuple<T...> &tpl) {
  rd_tuple(tpl);
}

template <size_t N = 0, typename T>
void rd(array<T, N> &x) {
  for (auto &d: x) rd(d);
}
template <class T>
void rd(vc<T> &x) {
  for (auto &d: x) rd(d);
}

void read() {}
template <class H, class... T>
void read(H &h, T &... t) {
  rd(h), read(t...);
}

void wt(const char c) {
  if (por == SZ) flush();
  obuf[por++] = c;
}
void wt(const string s) {
  for (char c: s) wt(c);
}
void wt(const char *s) {
  size_t len = strlen(s);
  for (size_t i = 0; i < len; i++) wt(s[i]);
}

template <typename T>
void wt_integer(T x) {
  if (por > SZ - 100) flush();
  if (x < 0) { obuf[por++] = '-', x = -x; }
  int outi;
  for (outi = 96; x >= 10000; outi -= 4) {
    memcpy(out + outi, pre.num[x % 10000], 4);
    x /= 10000;
  }
  if (x >= 1000) {
    memcpy(obuf + por, pre.num[x], 4);
    por += 4;
  } else if (x >= 100) {
    memcpy(obuf + por, pre.num[x] + 1, 3);
    por += 3;
  } else if (x >= 10) {
    int q = (x * 103) >> 10;
    obuf[por] = q | '0';
    obuf[por + 1] = (x - q * 10) | '0';
    por += 2;
  } else
    obuf[por++] = x | '0';
  memcpy(obuf + por, out + outi + 4, 96 - outi);
  por += 96 - outi;
}

template <typename T>
void wt_real(T x) {
  ostringstream oss;
  oss << fixed << setprecision(15) << double(x);
  string s = oss.str();
  wt(s);
}

void wt(int x) { wt_integer(x); }
void wt(ll x) { wt_integer(x); }
void wt(i128 x) { wt_integer(x); }
void wt(u32 x) { wt_integer(x); }
void wt(u64 x) { wt_integer(x); }
void wt(u128 x) { wt_integer(x); }
void wt(double x) { wt_real(x); }
void wt(long double x) { wt_real(x); }
void wt(f128 x) { wt_real(x); }

template <class T, class U>
void wt(const pair<T, U> val) {
  wt(val.first);
  wt(' ');
  wt(val.second);
}
template <size_t N = 0, typename T>
void wt_tuple(const T t) {
  if constexpr (N < std::tuple_size<T>::value) {
    if constexpr (N > 0) { wt(' '); }
    const auto x = std::get<N>(t);
    wt(x);
    wt_tuple<N + 1>(t);
  }
}
template <class... T>
void wt(tuple<T...> tpl) {
  wt_tuple(tpl);
}
template <class T, size_t S>
void wt(const array<T, S> val) {
  auto n = val.size();
  for (size_t i = 0; i < n; i++) {
    if (i) wt(' ');
    wt(val[i]);
  }
}
template <class T>
void wt(const vector<T> val) {
  auto n = val.size();
  for (size_t i = 0; i < n; i++) {
    if (i) wt(' ');
    wt(val[i]);
  }
}

void print() { wt('\n'); }
template <class Head, class... Tail>
void print(Head &&head, Tail &&... tail) {
  wt(head);
  if (sizeof...(Tail)) wt(' ');
  print(forward<Tail>(tail)...);
}

// gcc expansion. called automaticall after main.
void __attribute__((destructor)) _d() { flush(); }
} // namespace fastio
using fastio::read;
using fastio::print;
using fastio::flush;

#if defined(LOCAL)
#define SHOW(...) SHOW_IMPL(__VA_ARGS__, SHOW6, SHOW5, SHOW4, SHOW3, SHOW2, SHOW1)(__VA_ARGS__)
#define SHOW_IMPL(_1, _2, _3, _4, _5, _6, NAME, ...) NAME
#define SHOW1(x) print(#x, "=", (x)), flush()
#define SHOW2(x, y) print(#x, "=", (x), #y, "=", (y)), flush()
#define SHOW3(x, y, z) print(#x, "=", (x), #y, "=", (y), #z, "=", (z)), flush()
#define SHOW4(x, y, z, w) print(#x, "=", (x), #y, "=", (y), #z, "=", (z), #w, "=", (w)), flush()
#define SHOW5(x, y, z, w, v) print(#x, "=", (x), #y, "=", (y), #z, "=", (z), #w, "=", (w), #v, "=", (v)), flush()
#define SHOW6(x, y, z, w, v, u) print(#x, "=", (x), #y, "=", (y), #z, "=", (z), #w, "=", (w), #v, "=", (v), #u, "=", (u)), flush()
#else
#define SHOW(...)
#endif

#define INT(...)   \
  int __VA_ARGS__; \
  read(__VA_ARGS__)
#define LL(...)   \
  ll __VA_ARGS__; \
  read(__VA_ARGS__)
#define U32(...)   \
  u32 __VA_ARGS__; \
  read(__VA_ARGS__)
#define U64(...)   \
  u64 __VA_ARGS__; \
  read(__VA_ARGS__)
#define STR(...)      \
  string __VA_ARGS__; \
  read(__VA_ARGS__)
#define CHAR(...)   \
  char __VA_ARGS__; \
  read(__VA_ARGS__)
#define DBL(...)      \
  double __VA_ARGS__; \
  read(__VA_ARGS__)

#define VEC(type, name, size) \
  vector<type> name(size);    \
  read(name)
#define VV(type, name, h, w)                     \
  vector<vector<type>> name(h, vector<type>(w)); \
  read(name)

void YES(bool t = 1) { print(t ? "YES" : "NO"); }
void NO(bool t = 1) { YES(!t); }
void Yes(bool t = 1) { print(t ? "Yes" : "No"); }
void No(bool t = 1) { Yes(!t); }
void yes(bool t = 1) { print(t ? "yes" : "no"); }
void no(bool t = 1) { yes(!t); }
#line 3 "main.cpp"

#line 2 "/home/maspy/compro/library/ds/my_bitset.hpp"

// https://codeforces.com/contest/914/problem/F
// https://yukicoder.me/problems/no/142
// わずかに普通の bitset より遅いときもあるようだが，
// 固定長にしたくないときや slice 操作が必要なときに使う
struct My_Bitset {
  using T = My_Bitset;
  int N;
  vc<u64> dat;

  // x で埋める
  My_Bitset(int N = 0, int x = 0) : N(N) {
    assert(x == 0 || x == 1);
    u64 v = (x == 0 ? 0 : -1);
    dat.assign((N + 63) >> 6, v);
    if (N) dat.back() >>= (64 * len(dat) - N);
  }

  int size() { return N; }

  void resize(int size) {
    dat.resize((size + 63) >> 6);
    int remainingBits = size & 63;
    if (remainingBits != 0) {
      u64 mask = (u64(1) << remainingBits) - 1;
      dat.back() &= mask;
    }
    N = size;
  }

  static T from_string(string &S) {
    int N = len(S);
    T ANS(N);
    FOR(i, N) ANS[i] = (S[i] == '1');
    return ANS;
  }

  // thanks to chatgpt!
  class Proxy {
  public:
    Proxy(vc<u64> &d, int i) : dat(d), index(i) {}
    operator bool() const { return (dat[index >> 6] >> (index & 63)) & 1; }

    Proxy &operator=(u64 value) {
      dat[index >> 6] &= ~(u64(1) << (index & 63));
      dat[index >> 6] |= (value & 1) << (index & 63);
      return *this;
    }
    void flip() {
      dat[index >> 6] ^= (u64(1) << (index & 63)); // XOR to flip the bit
    }

  private:
    vc<u64> &dat;
    int index;
  };

  Proxy operator[](int i) { return Proxy(dat, i); }

  bool operator==(const T &p) {
    assert(N == p.N);
    FOR(i, len(dat)) if (dat[i] != p.dat[i]) return false;
    return true;
  }

  T &operator&=(const T &p) {
    assert(N == p.N);
    FOR(i, len(dat)) dat[i] &= p.dat[i];
    return *this;
  }
  T &operator|=(const T &p) {
    assert(N == p.N);
    FOR(i, len(dat)) dat[i] |= p.dat[i];
    return *this;
  }
  T &operator^=(const T &p) {
    assert(N == p.N);
    FOR(i, len(dat)) dat[i] ^= p.dat[i];
    return *this;
  }
  T operator&(const T &p) const { return T(*this) &= p; }
  T operator|(const T &p) const { return T(*this) |= p; }
  T operator^(const T &p) const { return T(*this) ^= p; }
  T operator~() const {
    T p = (*this);
    p.flip_range(0, N);
    return p;
  }

  int count() {
    int ans = 0;
    for (u64 val: dat) ans += popcnt(val);
    return ans;
  }

  int dot(T &p) {
    assert(N == p.N);
    int ans = 0;
    FOR(i, len(dat)) ans += popcnt(dat[i] & p.dat[i]);
    return ans;
  }

  int dot_mod_2(T &p) {
    assert(N == p.N);
    int ans = 0;
    FOR(i, len(dat)) ans ^= popcnt_mod_2(dat[i] & p.dat[i]);
    return ans;
  }

  int next(int i) {
    chmax(i, 0);
    if (i >= N) return N;
    int k = i >> 6;
    {
      u64 x = dat[k];
      int s = i & 63;
      x = (x >> s) << s;
      if (x) return (k << 6) | lowbit(x);
    }
    FOR(idx, k + 1, len(dat)) {
      if (dat[idx] == 0) continue;
      return (idx << 6) | lowbit(dat[idx]);
    }
    return N;
  }

  int prev(int i) {
    chmin(i, N - 1);
    if (i <= -1) return -1;
    int k = i >> 6;
    if ((i & 63) < 63) {
      u64 x = dat[k];
      x &= (u64(1) << ((i & 63) + 1)) - 1;
      if (x) return (k << 6) | topbit(x);
      --k;
    }
    FOR_R(idx, k + 1) {
      if (dat[idx] == 0) continue;
      return (idx << 6) | topbit(dat[idx]);
    }
    return -1;
  }

  My_Bitset range(int L, int R) {
    assert(L <= R);
    My_Bitset p(R - L);
    int rm = (R - L) & 63;
    FOR(rm) {
      p[R - L - 1] = bool((*this)[R - 1]);
      --R;
    }
    int n = (R - L) >> 6;
    int hi = L & 63;
    int lo = 64 - hi;
    int s = L >> 6;
    if (hi == 0) {
      FOR(i, n) { p.dat[i] ^= dat[s + i]; }
    } else {
      FOR(i, n) { p.dat[i] ^= (dat[s + i] >> hi) ^ (dat[s + i + 1] << lo); }
    }
    return p;
  }

  My_Bitset slice(int L, int R) { return range(L, R); }

  int count_range(int L, int R) {
    assert(L <= R);
    int cnt = 0;
    while ((L < R) && (L & 63)) cnt += (*this)[L++];
    while ((L < R) && (R & 63)) cnt += (*this)[--R];
    int l = L >> 6, r = R >> 6;
    FOR(i, l, r) cnt += popcnt(dat[i]);
    return cnt;
  }

  // [L,R) に p を代入
  void assign_to_range(int L, int R, My_Bitset &p) {
    assert(p.N == R - L);
    int a = 0, b = p.N;
    while (L < R && (L & 63)) { (*this)[L++] = bool(p[a++]); }
    while (L < R && (R & 63)) { (*this)[--R] = bool(p[--b]); }
    // p[a:b] を [L:R] に
    int l = L >> 6, r = R >> 6;
    int s = a >> 6, t = b >> t;
    int n = r - l;
    if (!(a & 63)) {
      FOR(i, n) dat[l + i] = p.dat[s + i];
    } else {
      int hi = a & 63;
      int lo = 64 - hi;
      FOR(i, n) dat[l + i] = (p.dat[s + i] >> hi) | (p.dat[1 + s + i] << lo);
    }
  }

  // [L,R) に p を xor
  void xor_to_range(int L, int R, My_Bitset &p) {
    assert(p.N == R - L);
    int a = 0, b = p.N;
    while (L < R && (L & 63)) {
      dat[L >> 6] ^= u64(p[a]) << (L & 63);
      ++a, ++L;
    }
    while (L < R && (R & 63)) {
      --b, --R;
      dat[R >> 6] ^= u64(p[b]) << (R & 63);
    }
    // p[a:b] を [L:R] に
    int l = L >> 6, r = R >> 6;
    int s = a >> 6, t = b >> t;
    int n = r - l;
    if (!(a & 63)) {
      FOR(i, n) dat[l + i] ^= p.dat[s + i];
    } else {
      int hi = a & 63;
      int lo = 64 - hi;
      FOR(i, n) dat[l + i] ^= (p.dat[s + i] >> hi) | (p.dat[1 + s + i] << lo);
    }
  }

  // 行列基本変形で使うやつ
  // p は [i:N) にしかないとして p を xor する
  void xor_suffix(int i, My_Bitset &p) {
    assert(N == p.N && 0 <= i && i < N);
    FOR(k, i / 64, len(dat)) { dat[k] ^= p.dat[k]; }
  }

  // [L,R) に p を and
  void and_to_range(int L, int R, My_Bitset &p) {
    assert(p.N == R - L);
    int a = 0, b = p.N;
    while (L < R && (L & 63)) {
      if (!p[a]) (*this)[L] = 0;
      a++, L++;
    }
    while (L < R && (R & 63)) {
      --b, --R;
      if (!p[b]) (*this)[R] = 0;
    }
    // p[a:b] を [L:R] に
    int l = L >> 6, r = R >> 6;
    int s = a >> 6, t = b >> t;
    int n = r - l;
    if (!(a & 63)) {
      FOR(i, n) dat[l + i] &= p.dat[s + i];
    } else {
      int hi = a & 63;
      int lo = 64 - hi;
      FOR(i, n) dat[l + i] &= (p.dat[s + i] >> hi) | (p.dat[1 + s + i] << lo);
    }
  }

  // [L,R) に p を or
  void or_to_range(int L, int R, My_Bitset &p) {
    assert(p.N == R - L);
    int a = 0, b = p.N;
    while (L < R && (L & 63)) {
      dat[L >> 6] |= u64(p[a]) << (L & 63);
      ++a, ++L;
    }
    while (L < R && (R & 63)) {
      --b, --R;
      dat[R >> 6] |= u64(p[b]) << (R & 63);
    }
    // p[a:b] を [L:R] に
    int l = L >> 6, r = R >> 6;
    int s = a >> 6, t = b >> t;
    int n = r - l;
    if (!(a & 63)) {
      FOR(i, n) dat[l + i] |= p.dat[s + i];
    } else {
      int hi = a & 63;
      int lo = 64 - hi;
      FOR(i, n) dat[l + i] |= (p.dat[s + i] >> hi) | (p.dat[1 + s + i] << lo);
    }
  }
  // 行列基本変形で使うやつ
  // p は [i:N) にしかないとして p を or する
  void or_suffix(int i, My_Bitset &p) {
    assert(N == p.N && 0 <= i && i < N);
    FOR(k, i / 64, len(dat)) { dat[k] |= p.dat[k]; }
  }

  // [L,R) を 1 に変更
  void set_range(int L, int R) {
    while (L < R && (L & 63)) { set(L++); }
    while (L < R && (R & 63)) { set(--R); }
    FOR(i, L >> 6, R >> 6) dat[i] = u64(-1);
  }

  // [L,R) を 1 に変更
  void reset_range(int L, int R) {
    while (L < R && (L & 63)) { reset(L++); }
    while (L < R && (R & 63)) { reset(--R); }
    FOR(i, L >> 6, R >> 6) dat[i] = u64(0);
  }

  // [L,R) を flip
  void flip_range(int L, int R) {
    while (L < R && (L & 63)) { flip(L++); }
    while (L < R && (R & 63)) { flip(--R); }
    FOR(i, L >> 6, R >> 6) dat[i] ^= u64(-1);
  }

  // bitset に仕様を合わせる
  void set(int i) { (*this)[i] = 1; }
  void reset(int i) { (*this)[i] = 0; }
  void flip(int i) { (*this)[i].flip(); }
  void set() {
    fill(all(dat), u64(-1));
    resize(N);
  }
  void reset() { fill(all(dat), 0); }
  void flip() {
    FOR(i, len(dat) - 1) { dat[i] = u64(-1) ^ dat[i]; }
    int i = len(dat) - 1;
    FOR(k, 64) {
      if (64 * i + k >= size()) break;
      flip(64 * i + k);
    }
  }
  bool any() {
    FOR(i, len(dat)) {
      if (dat[i]) return true;
    }
    return false;
  }

  bool ALL() {
    dat.resize((N + 63) >> 6);
    int r = N & 63;
    if (r != 0) {
      u64 mask = (u64(1) << r) - 1;
      if (dat.back() != mask) return 0;
    }
    for (int i = 0; i < N / 64; ++i)
      if (dat[i] != u64(-1)) return false;
    return true;
  }

  int _Find_first() { return next(0); }
  int _Find_next(int p) { return next(p + 1); }

  static string TO_STR[256];
  string to_string() const {
    if (TO_STR[0].empty()) precompute();
    string S;
    for (auto &x: dat) { FOR(i, 8) S += TO_STR[(x >> (8 * i) & 255)]; }
    S.resize(N);
    return S;
  }

  static void precompute() {
    FOR(s, 256) {
      string x;
      FOR(i, 8) x += '0' + (s >> i & 1);
      TO_STR[s] = x;
    }
  }
};
string My_Bitset::TO_STR[256];
#line 2 "/home/maspy/compro/library/string/suffix_array.hpp"

#line 2 "/home/maspy/compro/library/alg/monoid/min.hpp"

template <typename E>
struct Monoid_Min {
  using X = E;
  using value_type = X;
  static constexpr X op(const X &x, const X &y) noexcept { return min(x, y); }
  static constexpr X unit() { return infty<E>; }
  static constexpr bool commute = true;
};
#line 2 "/home/maspy/compro/library/ds/sparse_table/sparse_table.hpp"

// 冪等なモノイドであることを仮定。disjoint sparse table より x 倍高速
template <class Monoid>
struct Sparse_Table {
  using MX = Monoid;
  using X = typename MX::value_type;
  int n, log;
  vvc<X> dat;

  Sparse_Table() {}
  Sparse_Table(int n) { build(n); }
  template <typename F>
  Sparse_Table(int n, F f) {
    build(n, f);
  }
  Sparse_Table(const vc<X>& v) { build(v); }

  void build(int m) {
    build(m, [](int i) -> X { return MX::unit(); });
  }
  void build(const vc<X>& v) {
    build(len(v), [&](int i) -> X { return v[i]; });
  }
  template <typename F>
  void build(int m, F f) {
    n = m, log = 1;
    while ((1 << log) < n) ++log;
    dat.resize(log);
    dat[0].resize(n);
    FOR(i, n) dat[0][i] = f(i);

    FOR(i, log - 1) {
      dat[i + 1].resize(len(dat[i]) - (1 << i));
      FOR(j, len(dat[i]) - (1 << i)) {
        dat[i + 1][j] = MX::op(dat[i][j], dat[i][j + (1 << i)]);
      }
    }
  }

  X prod(int L, int R) {
    if (L == R) return MX::unit();
    if (R == L + 1) return dat[0][L];
    int k = topbit(R - L - 1);
    return MX::op(dat[k][L], dat[k][R - (1 << k)]);
  }

  template <class F>
  int max_right(const F check, int L) {
    assert(0 <= L && L <= n && check(MX::unit()));
    if (L == n) return n;
    int ok = L, ng = n + 1;
    while (ok + 1 < ng) {
      int k = (ok + ng) / 2;
      bool bl = check(prod(L, k));
      if (bl) ok = k;
      if (!bl) ng = k;
    }
    return ok;
  }

  template <class F>
  int min_left(const F check, int R) {
    assert(0 <= R && R <= n && check(MX::unit()));
    if (R == 0) return 0;
    int ok = R, ng = -1;
    while (ng + 1 < ok) {
      int k = (ok + ng) / 2;
      bool bl = check(prod(k, R));
      if (bl) ok = k;
      if (!bl) ng = k;
    }
    return ok;
  }
};
#line 5 "/home/maspy/compro/library/string/suffix_array.hpp"

// 辞書順 i 番目の suffix が j 文字目始まりであるとき、
// SA[i] = j, ISA[j] = i
// |S|>0 を前提（そうでない場合 dummy 文字を追加して利用せよ）
struct Suffix_Array {
  vc<int> SA;
  vc<int> ISA;
  vc<int> LCP;
  Sparse_Table<Monoid_Min<int>> seg;
  bool build_seg;

  Suffix_Array(string& s) {
    build_seg = 0;
    assert(len(s) > 0);
    char first = 127, last = 0;
    for (auto&& c: s) {
      chmin(first, c);
      chmax(last, c);
    }
    SA = calc_suffix_array(s, first, last);
    calc_LCP(s);
  }

  Suffix_Array(vc<int>& s) {
    build_seg = 0;
    assert(len(s) > 0);
    SA = calc_suffix_array(s);
    calc_LCP(s);
  }

  // lcp(S[i:], S[j:])
  int lcp(int i, int j) {
    if (!build_seg) {
      build_seg = true;
      seg.build(LCP);
    }
    int n = len(SA);
    if (i == n || j == n) return 0;
    if (i == j) return n - i;
    i = ISA[i], j = ISA[j];
    if (i > j) swap(i, j);
    return seg.prod(i, j);
  }

  // S[i:] との lcp が n 以上であるような半開区間
  pair<int, int> lcp_range(int i, int n) {
    if (!build_seg) {
      build_seg = true;
      seg.build(LCP);
    }
    i = ISA[i];
    int a = seg.min_left([&](auto e) -> bool { return e >= n; }, i);
    int b = seg.max_right([&](auto e) -> bool { return e >= n; }, i);
    return {a, b + 1};
  }

  // -1: S[L1:R1) < S[L2, R2)
  //  0: S[L1:R1) = S[L2, R2)
  // +1: S[L1:R1) > S[L2, R2)
  int compare(int L1, int R1, int L2, int R2) {
    int n1 = R1 - L1, n2 = R2 - L2;
    int n = lcp(L1, L2);
    if (n == n1 && n == n2) return 0;
    if (n == n1) return -1;
    if (n == n2) return 1;
    return (ISA[L1 + n] > ISA[L2 + n] ? 1 : -1);
  }

private:
  void induced_sort(const vc<int>& vect, int val_range, vc<int>& SA,
                    const vc<bool>& sl, const vc<int>& lms_idx) {
    vc<int> l(val_range, 0), r(val_range, 0);
    for (int c: vect) {
      if (c + 1 < val_range) ++l[c + 1];
      ++r[c];
    }
    partial_sum(l.begin(), l.end(), l.begin());
    partial_sum(r.begin(), r.end(), r.begin());
    fill(SA.begin(), SA.end(), -1);
    for (int i = (int)lms_idx.size() - 1; i >= 0; --i)
      SA[--r[vect[lms_idx[i]]]] = lms_idx[i];
    for (int i: SA)
      if (i >= 1 && sl[i - 1]) SA[l[vect[i - 1]]++] = i - 1;
    fill(r.begin(), r.end(), 0);
    for (int c: vect) ++r[c];
    partial_sum(r.begin(), r.end(), r.begin());
    for (int k = (int)SA.size() - 1, i = SA[k]; k >= 1; --k, i = SA[k])
      if (i >= 1 && !sl[i - 1]) { SA[--r[vect[i - 1]]] = i - 1; }
  }

  vc<int> SA_IS(const vc<int>& vect, int val_range) {
    const int n = vect.size();
    vc<int> SA(n), lms_idx;
    vc<bool> sl(n);
    sl[n - 1] = false;
    for (int i = n - 2; i >= 0; --i) {
      sl[i] = (vect[i] > vect[i + 1] || (vect[i] == vect[i + 1] && sl[i + 1]));
      if (sl[i] && !sl[i + 1]) lms_idx.push_back(i + 1);
    }
    reverse(lms_idx.begin(), lms_idx.end());
    induced_sort(vect, val_range, SA, sl, lms_idx);
    vc<int> new_lms_idx(lms_idx.size()), lms_vec(lms_idx.size());
    for (int i = 0, k = 0; i < n; ++i)
      if (!sl[SA[i]] && SA[i] >= 1 && sl[SA[i] - 1]) {
        new_lms_idx[k++] = SA[i];
      }
    int cur = 0;
    SA[n - 1] = cur;
    for (size_t k = 1; k < new_lms_idx.size(); ++k) {
      int i = new_lms_idx[k - 1], j = new_lms_idx[k];
      if (vect[i] != vect[j]) {
        SA[j] = ++cur;
        continue;
      }
      bool flag = false;
      for (int a = i + 1, b = j + 1;; ++a, ++b) {
        if (vect[a] != vect[b]) {
          flag = true;
          break;
        }
        if ((!sl[a] && sl[a - 1]) || (!sl[b] && sl[b - 1])) {
          flag = !((!sl[a] && sl[a - 1]) && (!sl[b] && sl[b - 1]));
          break;
        }
      }
      SA[j] = (flag ? ++cur : cur);
    }
    for (size_t i = 0; i < lms_idx.size(); ++i) lms_vec[i] = SA[lms_idx[i]];
    if (cur + 1 < (int)lms_idx.size()) {
      auto lms_SA = SA_IS(lms_vec, cur + 1);
      for (size_t i = 0; i < lms_idx.size(); ++i) {
        new_lms_idx[i] = lms_idx[lms_SA[i]];
      }
    }
    induced_sort(vect, val_range, SA, sl, new_lms_idx);
    return SA;
  }

  vc<int> calc_suffix_array(const string& s, const char first = 'a',
                            const char last = 'z') {
    vc<int> vect(s.size() + 1);
    copy(begin(s), end(s), begin(vect));
    for (auto& x: vect) x -= (int)first - 1;
    vect.back() = 0;
    auto ret = SA_IS(vect, (int)last - (int)first + 2);
    ret.erase(ret.begin());
    return ret;
  }

  vc<int> calc_suffix_array(const vc<int>& s) {
    vc<int> ss = s;
    UNIQUE(ss);

    vc<int> vect(s.size() + 1);
    copy(all(s), vect.begin());
    for (auto& x: vect) x = LB(ss, x) + 1;
    vect.back() = 0;
    auto ret = SA_IS(vect, MAX(vect) + 2);
    ret.erase(ret.begin());
    return ret;
  }

  template <typename STRING>
  void calc_LCP(const STRING& s) {
    int n = s.size(), k = 0;
    ISA.resize(n);
    LCP.resize(n);
    for (int i = 0; i < n; i++) ISA[SA[i]] = i;
    for (int i = 0; i < n; i++, k ? k-- : 0) {
      if (ISA[i] == n - 1) {
        k = 0;
        continue;
      }
      int j = SA[ISA[i] + 1];
      while (i + k < n && j + k < n && s[i + k] == s[j + k]) k++;
      LCP[ISA[i]] = k;
    }
    LCP.resize(n - 1);
  }
};
#line 1 "/home/maspy/compro/library/string/suffix_tree.hpp"

#line 1 "/home/maspy/compro/library/seq/cartesian_tree.hpp"
/*
辞書順で高さを unique して、木にしている。
極大長方形アルゴリズムで線形時間構築。
*/
template <typename T, bool IS_MIN>
struct CartesianTree {
  int n;
  vc<T>& A;
  vc<pair<int, int>> range;
  vc<int> lch, rch, par;
  int root;

  CartesianTree(vc<T>& A) : n(len(A)), A(A) {
    range.assign(n, {-1, -1});
    lch.assign(n, -1);
    rch.assign(n, -1);
    par.assign(n, -1);
    if (n == 1) {
      range[0] = {0, 1};
      root = 0;
      return;
    }
    auto is_sm = [&](int i, int j) -> bool {
      if (IS_MIN) return (A[i] < A[j]) || (A[i] == A[j] && i < j);
      return (A[i] > A[j]) || (A[i] == A[j] && i < j);
    };
    vc<int> st;
    FOR(i, n) {
      while (!st.empty() && is_sm(i, st.back())) {
        lch[i] = st.back();
        st.pop_back();
      }
      range[i].fi = (st.empty() ? 0 : st.back() + 1);
      st.eb(i);
    }
    st.clear();
    FOR_R(i, n) {
      while (!st.empty() && is_sm(i, st.back())) {
        rch[i] = st.back();
        st.pop_back();
      }
      range[i].se = (st.empty() ? n : st.back());
      st.eb(i);
    }
    FOR(i, n) if (lch[i] != -1) par[lch[i]] = i;
    FOR(i, n) if (rch[i] != -1) par[rch[i]] = i;
    FOR(i, n) if (par[i] == -1) root = i;
  }

  // (l, r, h)
  tuple<int, int, T> maximum_rectangle(int i) {
    auto [l, r] = range[i];
    return {l, r, A[i]};
  }

  // (l, r, h)
  T max_rectangle_area() {
    assert(IS_MIN);
    T res = 0;
    FOR(i, n) {
      auto [l, r, h] = maximum_rectangle(i);
      chmax(res, (r - l) * h);
    }
    return res;
  }

  ll count_subrectangle(bool baseline) {
    assert(IS_MIN);
    ll res = 0;
    FOR(i, n) {
      auto [l, r, h] = maximum_rectangle(i);
      ll x = (baseline ? h : h * (h + 1) / 2);
      res += x * (i - l + 1) * (r - i);
    }
    return res;
  }
};
#line 2 "/home/maspy/compro/library/graph/base.hpp"

template <typename T>
struct Edge {
  int frm, to;
  T cost;
  int id;
};

template <typename T = int, bool directed = false>
struct Graph {
  static constexpr bool is_directed = directed;
  int N, M;
  using cost_type = T;
  using edge_type = Edge<T>;
  vector<edge_type> edges;
  vector<int> indptr;
  vector<edge_type> csr_edges;
  vc<int> vc_deg, vc_indeg, vc_outdeg;
  bool prepared;

  class OutgoingEdges {
  public:
    OutgoingEdges(const Graph* G, int l, int r) : G(G), l(l), r(r) {}

    const edge_type* begin() const {
      if (l == r) { return 0; }
      return &G->csr_edges[l];
    }

    const edge_type* end() const {
      if (l == r) { return 0; }
      return &G->csr_edges[r];
    }

  private:
    const Graph* G;
    int l, r;
  };

  bool is_prepared() { return prepared; }

  Graph() : N(0), M(0), prepared(0) {}
  Graph(int N) : N(N), M(0), prepared(0) {}

  void build(int n) {
    N = n, M = 0;
    prepared = 0;
    edges.clear();
    indptr.clear();
    csr_edges.clear();
    vc_deg.clear();
    vc_indeg.clear();
    vc_outdeg.clear();
  }

  void add(int frm, int to, T cost = 1, int i = -1) {
    assert(!prepared);
    assert(0 <= frm && 0 <= to && to < N);
    if (i == -1) i = M;
    auto e = edge_type({frm, to, cost, i});
    edges.eb(e);
    ++M;
  }

#ifdef FASTIO
  // wt, off
  void read_tree(bool wt = false, int off = 1) { read_graph(N - 1, wt, off); }

  void read_graph(int M, bool wt = false, int off = 1) {
    for (int m = 0; m < M; ++m) {
      INT(a, b);
      a -= off, b -= off;
      if (!wt) {
        add(a, b);
      } else {
        T c;
        read(c);
        add(a, b, c);
      }
    }
    build();
  }
#endif

  void build() {
    assert(!prepared);
    prepared = true;
    indptr.assign(N + 1, 0);
    for (auto&& e: edges) {
      indptr[e.frm + 1]++;
      if (!directed) indptr[e.to + 1]++;
    }
    for (int v = 0; v < N; ++v) { indptr[v + 1] += indptr[v]; }
    auto counter = indptr;
    csr_edges.resize(indptr.back() + 1);
    for (auto&& e: edges) {
      csr_edges[counter[e.frm]++] = e;
      if (!directed)
        csr_edges[counter[e.to]++] = edge_type({e.to, e.frm, e.cost, e.id});
    }
  }

  OutgoingEdges operator[](int v) const {
    assert(prepared);
    return {this, indptr[v], indptr[v + 1]};
  }

  vc<int> deg_array() {
    if (vc_deg.empty()) calc_deg();
    return vc_deg;
  }

  pair<vc<int>, vc<int>> deg_array_inout() {
    if (vc_indeg.empty()) calc_deg_inout();
    return {vc_indeg, vc_outdeg};
  }

  int deg(int v) {
    if (vc_deg.empty()) calc_deg();
    return vc_deg[v];
  }

  int in_deg(int v) {
    if (vc_indeg.empty()) calc_deg_inout();
    return vc_indeg[v];
  }

  int out_deg(int v) {
    if (vc_outdeg.empty()) calc_deg_inout();
    return vc_outdeg[v];
  }

#ifdef FASTIO
  void debug() {
    print("Graph");
    if (!prepared) {
      print("frm to cost id");
      for (auto&& e: edges) print(e.frm, e.to, e.cost, e.id);
    } else {
      print("indptr", indptr);
      print("frm to cost id");
      FOR(v, N) for (auto&& e: (*this)[v]) print(e.frm, e.to, e.cost, e.id);
    }
  }
#endif

  vc<int> new_idx;
  vc<bool> used_e;

  // G における頂点 V[i] が、新しいグラフで i になるようにする
  // {G, es}
  // sum(deg(v)) の計算量になっていて、
  // 新しいグラフの n+m より大きい可能性があるので注意
  Graph<T, directed> rearrange(vc<int> V, bool keep_eid = 0) {
    if (len(new_idx) != N) new_idx.assign(N, -1);
    int n = len(V);
    FOR(i, n) new_idx[V[i]] = i;
    Graph<T, directed> G(n);
    vc<int> history;
    FOR(i, n) {
      for (auto&& e: (*this)[V[i]]) {
        if (len(used_e) <= e.id) used_e.resize(e.id + 1);
        if (used_e[e.id]) continue;
        int a = e.frm, b = e.to;
        if (new_idx[a] != -1 && new_idx[b] != -1) {
          history.eb(e.id);
          used_e[e.id] = 1;
          int eid = (keep_eid ? e.id : -1);
          G.add(new_idx[a], new_idx[b], e.cost, eid);
        }
      }
    }
    FOR(i, n) new_idx[V[i]] = -1;
    for (auto&& eid: history) used_e[eid] = 0;
    G.build();
    return G;
  }

  Graph<T, true> to_directed_tree(int root = -1) {
    if (root == -1) root = 0;
    assert(!is_directed && prepared && M == N - 1);
    Graph<T, true> G1(N);
    vc<int> par(N, -1);
    auto dfs = [&](auto& dfs, int v) -> void {
      for (auto& e: (*this)[v]) {
        if (e.to == par[v]) continue;
        par[e.to] = v, dfs(dfs, e.to);
      }
    };
    dfs(dfs, root);
    for (auto& e: edges) {
      int a = e.frm, b = e.to;
      if (par[a] == b) swap(a, b);
      assert(par[b] == a);
      G1.add(a, b, e.cost);
    }
    G1.build();
    return G1;
  }

private:
  void calc_deg() {
    assert(vc_deg.empty());
    vc_deg.resize(N);
    for (auto&& e: edges) vc_deg[e.frm]++, vc_deg[e.to]++;
  }

  void calc_deg_inout() {
    assert(vc_indeg.empty());
    vc_indeg.resize(N);
    vc_outdeg.resize(N);
    for (auto&& e: edges) { vc_indeg[e.to]++, vc_outdeg[e.frm]++; }
  }
};
#line 5 "/home/maspy/compro/library/string/suffix_tree.hpp"

// https://twitter.com/maspy_stars/status/1565901414236205057?s=20&t=S2Tu6ayozHcakxai8dmh4g
// 各ノードは、suffix array での長方形領域と見なして、
// グラフおよび、領域データを作る。
// sample: test/my_test/suffix_tree.test.cpp
template <typename STRING>
struct Suffix_Tree {
  STRING& S;
  Suffix_Array& X;
  Suffix_Tree(STRING& S, Suffix_Array& X) : S(S), X(X) {}

  pair<Graph<int, 1>, vc<tuple<int, int, int, int>>> build() {
    auto SA = X.SA;
    auto ISA = X.ISA;
    auto LCP = X.LCP;

    vc<tuple<int, int, int, int>> dat;
    vc<pair<int, int>> edges;

    int N = len(SA);
    if (N == 1) {
      Graph<int, 1> G(2);
      G.add(0, 1);
      G.build();
      dat.eb(0, 1, 0, 1), dat.eb(0, 1, 1, 2);
      return {G, dat};
    }

    dat.eb(0, N, -1, 0);
    CartesianTree<int, true> CT(LCP);

    auto dfs = [&](auto& dfs, int p, int idx, int h) -> void {
      int L = CT.range[idx].fi;
      int R = CT.range[idx].se + 1;
      int hh = LCP[idx];
      if (h < hh) {
        edges.eb(p, len(dat));
        p = len(dat);
        dat.eb(L, R, h, hh);
      }
      if (CT.lch[idx] == -1) {
        if (hh < N - SA[idx]) {
          edges.eb(p, len(dat));
          dat.eb(idx, idx + 1, hh, N - SA[idx]);
        }
      } else {
        dfs(dfs, p, CT.lch[idx], hh);
      }
      if (CT.rch[idx] == -1) {
        if (hh < N - SA[idx + 1]) {
          edges.eb(p, len(dat));
          dat.eb(idx + 1, idx + 2, hh, N - SA[idx + 1]);
        }
      } else {
        dfs(dfs, p, CT.rch[idx], hh);
      }
    };
    int r = CT.root;
    if (LCP[r] > 0) {
      edges.eb(0, 1);
      dat.eb(0, N, 0, LCP[r]);
      dfs(dfs, 1, r, LCP[r]);
    } else {
      dfs(dfs, 0, r, 0);
    }
    for (auto& [a, b, c, d]: dat) ++c, ++d;

    Graph<int, 1> G(len(dat));
    for (auto&& [a, b]: edges) G.add(a, b);
    G.build();
    return {G, dat};
  }

  // trie の要領ですすむ（failure link はない）
  // (node, length)
  // 行き過ぎ：(-1,0)
  pair<int, int> next(Graph<int, 1>& G, vc<tuple<int, int, int, int>>& dat, pair<int, int> p, int ch) {
    auto [node, length] = p;
    if (node == -1) return {-1, 0};
    auto [l, r, a, b] = dat[node];
    if (length + 1 < b) {
      int i = X.SA[l];
      // S[i:i+length]
      if (ch != S[i + length]) return {-1, 0};
      return {node, length + 1};
    }
    for (auto& e: G[node]) {
      int n = e.to;
      auto [l, r, a, b] = dat[n];
      assert(a == length + 1);
      int i = X.SA[l];
      // S[i:i+length]
      if (ch == S[i + length]) return {n, length + 1};
    }
    return {-1, 0};
  }
};
#line 2 "/home/maspy/compro/library/nt/primetable.hpp"

template <typename T = int>
vc<T> primetable(int LIM) {
  ++LIM;
  const int S = 32768;
  static int done = 2;
  static vc<T> primes = {2}, sieve(S + 1);

  if (done < LIM) {
    done = LIM;

    primes = {2}, sieve.assign(S + 1, 0);
    const int R = LIM / 2;
    primes.reserve(int(LIM / log(LIM) * 1.1));
    vc<pair<int, int>> cp;
    for (int i = 3; i <= S; i += 2) {
      if (!sieve[i]) {
        cp.eb(i, i * i / 2);
        for (int j = i * i; j <= S; j += 2 * i) sieve[j] = 1;
      }
    }
    for (int L = 1; L <= R; L += S) {
      array<bool, S> block{};
      for (auto& [p, idx]: cp)
        for (int i = idx; i < S + L; idx = (i += p)) block[i - L] = 1;
      FOR(i, min(S, R - L)) if (!block[i]) primes.eb((L + i) * 2 + 1);
    }
  }
  int k = LB(primes, LIM + 1);
  return {primes.begin(), primes.begin() + k};
}
#line 3 "/home/maspy/compro/library/nt/lpf_table.hpp"

// [0, LIM], 0, 1 には -1 が入る。
vc<int> lpf_table(ll LIM) {
  auto primes = primetable(LIM);
  vc<int> res(LIM + 1, -1);
  FOR_R(i, len(primes)) {
    auto p = primes[i];
    FOR3(j, 1, LIM / p + 1) res[p * j] = p;
  }
  return res;
}
#line 2 "/home/maspy/compro/library/ds/fastset.hpp"

// 64-ary tree
// space: (N/63) * u64
struct FastSet {
  static constexpr u32 B = 64;
  int n, log;
  vvc<u64> seg;

  FastSet() {}
  FastSet(int n) { build(n); }

  int size() { return n; }

  template <typename F>
  FastSet(int n, F f) {
    build(n, f);
  }

  void build(int m) {
    seg.clear();
    n = m;
    do {
      seg.push_back(vc<u64>((m + B - 1) / B));
      m = (m + B - 1) / B;
    } while (m > 1);
    log = len(seg);
  }
  template <typename F>
  void build(int n, F f) {
    build(n);
    FOR(i, n) { seg[0][i / B] |= u64(f(i)) << (i % B); }
    FOR(h, log - 1) {
      FOR(i, len(seg[h])) {
        seg[h + 1][i / B] |= u64(bool(seg[h][i])) << (i % B);
      }
    }
  }

  bool operator[](int i) const { return seg[0][i / B] >> (i % B) & 1; }
  void insert(int i) {
    for (int h = 0; h < log; h++) {
      seg[h][i / B] |= u64(1) << (i % B), i /= B;
    }
  }
  void add(int i) { insert(i); }
  void erase(int i) {
    u64 x = 0;
    for (int h = 0; h < log; h++) {
      seg[h][i / B] &= ~(u64(1) << (i % B));
      seg[h][i / B] |= x << (i % B);
      x = bool(seg[h][i / B]);
      i /= B;
    }
  }
  void remove(int i) { erase(i); }

  // min[x,n) or n
  int next(int i) {
    assert(i <= n);
    chmax(i, 0);
    for (int h = 0; h < log; h++) {
      if (i / B == seg[h].size()) break;
      u64 d = seg[h][i / B] >> (i % B);
      if (!d) {
        i = i / B + 1;
        continue;
      }
      i += lowbit(d);
      for (int g = h - 1; g >= 0; g--) {
        i *= B;
        i += lowbit(seg[g][i / B]);
      }
      return i;
    }
    return n;
  }

  // max [0,x], or -1
  int prev(int i) {
    assert(i >= -1);
    if (i >= n) i = n - 1;
    for (int h = 0; h < log; h++) {
      if (i == -1) break;
      u64 d = seg[h][i / B] << (63 - i % B);
      if (!d) {
        i = i / B - 1;
        continue;
      }
      i -= __builtin_clzll(d);
      for (int g = h - 1; g >= 0; g--) {
        i *= B;
        i += topbit(seg[g][i / B]);
      }
      return i;
    }
    return -1;
  }

  bool any(int l, int r) { return next(l) < r; }

  // [l, r)
  template <typename F>
  void enumerate(int l, int r, F f) {
    for (int x = next(l); x < r; x = next(x + 1)) f(x);
  }

  string to_string() {
    string s(n, '?');
    for (int i = 0; i < n; ++i) s[i] = ((*this)[i] ? '1' : '0');
    return s;
  }
};
#line 2 "/home/maspy/compro/library/nt/factor.hpp"

#line 2 "/home/maspy/compro/library/random/base.hpp"

u64 RNG_64() {
  static uint64_t x_
      = uint64_t(chrono::duration_cast<chrono::nanoseconds>(chrono::high_resolution_clock::now().time_since_epoch()).count()) * 10150724397891781847ULL;
  x_ ^= x_ << 7;
  return x_ ^= x_ >> 9;
}

u64 RNG(u64 lim) { return RNG_64() % lim; }

ll RNG(ll l, ll r) { return l + RNG_64() % (r - l); }
#line 2 "/home/maspy/compro/library/mod/mongomery_modint.hpp"

// odd mod.
// x の代わりに rx を持つ
template <int id, typename U1, typename U2>
struct Mongomery_modint {
  using mint = Mongomery_modint;
  inline static U1 m, r, n2;
  static constexpr int W = numeric_limits<U1>::digits;

  static void set_mod(U1 mod) {
    assert(mod & 1 && mod <= U1(1) << (W - 2));
    m = mod, n2 = -U2(m) % m, r = m;
    FOR(5) r *= 2 - m * r;
    r = -r;
    assert(r * m == U1(-1));
  }
  static U1 reduce(U2 b) { return (b + U2(U1(b) * r) * m) >> W; }

  U1 x;
  Mongomery_modint() : x(0) {}
  Mongomery_modint(U1 x) : x(reduce(U2(x) * n2)){};
  U1 val() const {
    U1 y = reduce(x);
    return y >= m ? y - m : y;
  }
  mint &operator+=(mint y) {
    x = ((x += y.x) >= m ? x - m : x);
    return *this;
  }
  mint &operator-=(mint y) {
    x -= (x >= y.x ? y.x : y.x - m);
    return *this;
  }
  mint &operator*=(mint y) {
    x = reduce(U2(x) * y.x);
    return *this;
  }
  mint operator+(mint y) const { return mint(*this) += y; }
  mint operator-(mint y) const { return mint(*this) -= y; }
  mint operator*(mint y) const { return mint(*this) *= y; }
  bool operator==(mint y) const {
    return (x >= m ? x - m : x) == (y.x >= m ? y.x - m : y.x);
  }
  bool operator!=(mint y) const { return not operator==(y); }
  mint pow(ll n) const {
    assert(n >= 0);
    mint y = 1, z = *this;
    for (; n; n >>= 1, z *= z)
      if (n & 1) y *= z;
    return y;
  }
};

template <int id>
using Mongomery_modint_32 = Mongomery_modint<id, u32, u64>;
template <int id>
using Mongomery_modint_64 = Mongomery_modint<id, u64, u128>;
#line 3 "/home/maspy/compro/library/nt/primetest.hpp"

bool primetest(const u64 x) {
  assert(x < u64(1) << 62);
  if (x == 2 or x == 3 or x == 5 or x == 7) return true;
  if (x % 2 == 0 or x % 3 == 0 or x % 5 == 0 or x % 7 == 0) return false;
  if (x < 121) return x > 1;
  const u64 d = (x - 1) >> lowbit(x - 1);

  using mint = Mongomery_modint_64<202311020>;

  mint::set_mod(x);
  const mint one(u64(1)), minus_one(x - 1);
  auto ok = [&](u64 a) -> bool {
    auto y = mint(a).pow(d);
    u64 t = d;
    while (y != one && y != minus_one && t != x - 1) y *= y, t <<= 1;
    if (y != minus_one && t % 2 == 0) return false;
    return true;
  };
  if (x < (u64(1) << 32)) {
    for (u64 a: {2, 7, 61})
      if (!ok(a)) return false;
  } else {
    for (u64 a: {2, 325, 9375, 28178, 450775, 9780504, 1795265022}) {
      if (!ok(a)) return false;
    }
  }
  return true;
}
#line 5 "/home/maspy/compro/library/nt/factor.hpp"

template <typename mint>
ll rho(ll n, ll c) {
  assert(n > 1);
  const mint cc(c);
  auto f = [&](mint x) { return x * x + cc; };
  mint x = 1, y = 2, z = 1, q = 1;
  ll g = 1;
  const ll m = 1LL << (__lg(n) / 5);
  for (ll r = 1; g == 1; r <<= 1) {
    x = y;
    FOR(r) y = f(y);
    for (ll k = 0; k < r && g == 1; k += m) {
      z = y;
      FOR(min(m, r - k)) y = f(y), q *= x - y;
      g = gcd(q.val(), n);
    }
  }
  if (g == n) do {
      z = f(z);
      g = gcd((x - z).val(), n);
    } while (g == 1);
  return g;
}

ll find_prime_factor(ll n) {
  assert(n > 1);
  if (primetest(n)) return n;
  FOR(100) {
    ll m = 0;
    if (n < (1 << 30)) {
      using mint = Mongomery_modint_32<20231025>;
      mint::set_mod(n);
      m = rho<mint>(n, RNG(0, n));
    } else {
      using mint = Mongomery_modint_64<20231025>;
      mint::set_mod(n);
      m = rho<mint>(n, RNG(0, n));
    }
    if (primetest(m)) return m;
    n = m;
  }
  assert(0);
  return -1;
}

// ソートしてくれる
vc<pair<ll, int>> factor(ll n) {
  assert(n >= 1);
  vc<pair<ll, int>> pf;
  FOR(p, 2, 100) {
    if (p * p > n) break;
    if (n % p == 0) {
      ll e = 0;
      do { n /= p, e += 1; } while (n % p == 0);
      pf.eb(p, e);
    }
  }
  while (n > 1) {
    ll p = find_prime_factor(n);
    ll e = 0;
    do { n /= p, e += 1; } while (n % p == 0);
    pf.eb(p, e);
  }
  sort(all(pf));
  return pf;
}

vc<pair<ll, int>> factor_by_lpf(ll n, vc<int>& lpf) {
  vc<pair<ll, int>> res;
  while (n > 1) {
    int p = lpf[n];
    int e = 0;
    while (n % p == 0) {
      n /= p;
      ++e;
    }
    res.eb(p, e);
  }
  return res;
}
#line 2 "/home/maspy/compro/library/graph/tree.hpp"

#line 4 "/home/maspy/compro/library/graph/tree.hpp"

// HLD euler tour をとっていろいろ。
template <typename GT>
struct Tree {
  using Graph_type = GT;
  GT &G;
  using WT = typename GT::cost_type;
  int N;
  vector<int> LID, RID, head, V, parent, VtoE;
  vc<int> depth;
  vc<WT> depth_weighted;

  Tree(GT &G, int r = 0, bool hld = 1) : G(G) { build(r, hld); }

  void build(int r = 0, bool hld = 1) {
    if (r == -1) return; // build を遅延したいとき
    N = G.N;
    LID.assign(N, -1), RID.assign(N, -1), head.assign(N, r);
    V.assign(N, -1), parent.assign(N, -1), VtoE.assign(N, -1);
    depth.assign(N, -1), depth_weighted.assign(N, 0);
    assert(G.is_prepared());
    int t1 = 0;
    dfs_sz(r, -1, hld);
    dfs_hld(r, t1);
  }

  void dfs_sz(int v, int p, bool hld) {
    auto &sz = RID;
    parent[v] = p;
    depth[v] = (p == -1 ? 0 : depth[p] + 1);
    sz[v] = 1;
    int l = G.indptr[v], r = G.indptr[v + 1];
    auto &csr = G.csr_edges;
    // 使う辺があれば先頭にする
    for (int i = r - 2; i >= l; --i) {
      if (hld && depth[csr[i + 1].to] == -1) swap(csr[i], csr[i + 1]);
    }
    int hld_sz = 0;
    for (int i = l; i < r; ++i) {
      auto e = csr[i];
      if (depth[e.to] != -1) continue;
      depth_weighted[e.to] = depth_weighted[v] + e.cost;
      VtoE[e.to] = e.id;
      dfs_sz(e.to, v, hld);
      sz[v] += sz[e.to];
      if (hld && chmax(hld_sz, sz[e.to]) && l < i) { swap(csr[l], csr[i]); }
    }
  }

  void dfs_hld(int v, int &times) {
    LID[v] = times++;
    RID[v] += LID[v];
    V[LID[v]] = v;
    bool heavy = true;
    for (auto &&e: G[v]) {
      if (depth[e.to] <= depth[v]) continue;
      head[e.to] = (heavy ? head[v] : e.to);
      heavy = false;
      dfs_hld(e.to, times);
    }
  }

  vc<int> heavy_path_at(int v) {
    vc<int> P = {v};
    while (1) {
      int a = P.back();
      for (auto &&e: G[a]) {
        if (e.to != parent[a] && head[e.to] == v) {
          P.eb(e.to);
          break;
        }
      }
      if (P.back() == a) break;
    }
    return P;
  }

  int heavy_child(int v) {
    int k = LID[v] + 1;
    if (k == N) return -1;
    int w = V[k];
    return (parent[w] == v ? w : -1);
  }

  int e_to_v(int eid) {
    auto e = G.edges[eid];
    return (parent[e.frm] == e.to ? e.frm : e.to);
  }
  int v_to_e(int v) { return VtoE[v]; }
  int get_eid(int u, int v) {
    if (parent[u] != v) swap(u, v);
    assert(parent[u] == v);
    return VtoE[u];
  }

  int ELID(int v) { return 2 * LID[v] - depth[v]; }
  int ERID(int v) { return 2 * RID[v] - depth[v] - 1; }

  // 目標地点へ進む個数が k
  int LA(int v, int k) {
    assert(k <= depth[v]);
    while (1) {
      int u = head[v];
      if (LID[v] - k >= LID[u]) return V[LID[v] - k];
      k -= LID[v] - LID[u] + 1;
      v = parent[u];
    }
  }
  int la(int u, int v) { return LA(u, v); }

  int LCA(int u, int v) {
    for (;; v = parent[head[v]]) {
      if (LID[u] > LID[v]) swap(u, v);
      if (head[u] == head[v]) return u;
    }
  }

  int meet(int a, int b, int c) { return LCA(a, b) ^ LCA(a, c) ^ LCA(b, c); }
  int lca(int u, int v) { return LCA(u, v); }

  int subtree_size(int v, int root = -1) {
    if (root == -1) return RID[v] - LID[v];
    if (v == root) return N;
    int x = jump(v, root, 1);
    if (in_subtree(v, x)) return RID[v] - LID[v];
    return N - RID[x] + LID[x];
  }

  int dist(int a, int b) {
    int c = LCA(a, b);
    return depth[a] + depth[b] - 2 * depth[c];
  }

  WT dist_weighted(int a, int b) {
    int c = LCA(a, b);
    return depth_weighted[a] + depth_weighted[b] - WT(2) * depth_weighted[c];
  }

  // a is in b
  bool in_subtree(int a, int b) { return LID[b] <= LID[a] && LID[a] < RID[b]; }

  int jump(int a, int b, ll k) {
    if (k == 1) {
      if (a == b) return -1;
      return (in_subtree(b, a) ? LA(b, depth[b] - depth[a] - 1) : parent[a]);
    }
    int c = LCA(a, b);
    int d_ac = depth[a] - depth[c];
    int d_bc = depth[b] - depth[c];
    if (k > d_ac + d_bc) return -1;
    if (k <= d_ac) return LA(a, k);
    return LA(b, d_ac + d_bc - k);
  }

  vc<int> collect_child(int v) {
    vc<int> res;
    for (auto &&e: G[v])
      if (e.to != parent[v]) res.eb(e.to);
    return res;
  }

  vc<int> collect_light(int v) {
    vc<int> res;
    bool skip = true;
    for (auto &&e: G[v])
      if (e.to != parent[v]) {
        if (!skip) res.eb(e.to);
        skip = false;
      }
    return res;
  }

  vc<pair<int, int>> get_path_decomposition(int u, int v, bool edge) {
    // [始点, 終点] の"閉"区間列。
    vc<pair<int, int>> up, down;
    while (1) {
      if (head[u] == head[v]) break;
      if (LID[u] < LID[v]) {
        down.eb(LID[head[v]], LID[v]);
        v = parent[head[v]];
      } else {
        up.eb(LID[u], LID[head[u]]);
        u = parent[head[u]];
      }
    }
    if (LID[u] < LID[v]) down.eb(LID[u] + edge, LID[v]);
    elif (LID[v] + edge <= LID[u]) up.eb(LID[u], LID[v] + edge);
    reverse(all(down));
    up.insert(up.end(), all(down));
    return up;
  }

  // 辺の列の情報 (frm,to,str)
  // str = "heavy_up", "heavy_down", "light_up", "light_down"
  vc<tuple<int, int, string>> get_path_decomposition_detail(int u, int v) {
    vc<tuple<int, int, string>> up, down;
    while (1) {
      if (head[u] == head[v]) break;
      if (LID[u] < LID[v]) {
        if (v != head[v]) down.eb(head[v], v, "heavy_down"), v = head[v];
        down.eb(parent[v], v, "light_down"), v = parent[v];
      } else {
        if (u != head[u]) up.eb(u, head[u], "heavy_up"), u = head[u];
        up.eb(u, parent[u], "light_up"), u = parent[u];
      }
    }
    if (LID[u] < LID[v]) down.eb(u, v, "heavy_down");
    elif (LID[v] < LID[u]) up.eb(u, v, "heavy_up");
    reverse(all(down));
    concat(up, down);
    return up;
  }

  vc<int> restore_path(int u, int v) {
    vc<int> P;
    for (auto &&[a, b]: get_path_decomposition(u, v, 0)) {
      if (a <= b) {
        FOR(i, a, b + 1) P.eb(V[i]);
      } else {
        FOR_R(i, b, a + 1) P.eb(V[i]);
      }
    }
    return P;
  }

  // path [a,b] と [c,d] の交わり. 空ならば {-1,-1}.
  // https://codeforces.com/problemset/problem/500/G
  pair<int, int> path_intersection(int a, int b, int c, int d) {
    int ab = lca(a, b), ac = lca(a, c), ad = lca(a, d);
    int bc = lca(b, c), bd = lca(b, d), cd = lca(c, d);
    int x = ab ^ ac ^ bc, y = ab ^ ad ^ bd; // meet(a,b,c), meet(a,b,d)
    if (x != y) return {x, y};
    int z = ac ^ ad ^ cd;
    if (x != z) x = -1;
    return {x, x};
  }
};
#line 1 "/home/maspy/compro/library/ds/sum_over_bit_positions.hpp"

// https://qoj.ac/contest/1784/problem/9244
// sum bitset[i]*wt[i]
// T は 11bit sum がおさまれば ok
template <typename T, int MAXSIZE>
struct Sum_Over_Bit_Positions {
  int N;
  vc<T> base;
  static T table[MAXSIZE / 64 * 6 + 10][1 << 11];

  template <typename F>
  Sum_Over_Bit_Positions(int N, F f) : N(N) {
    base.resize(N);
    assert(0 <= N && N < MAXSIZE);
    int NB = (N + 63) / 64;
    FOR(block, NB) {
      FOR(k, 6) {
        int b = 6 * block + k;
        FOR(i, 11) {
          int idx = 64 * block + 11 * k + i;
          T x = 0;
          if ((k < 5 || i < 9) && idx < N) x = base[idx] = f(idx);
          FOR(s, 1 << i) { table[b][s | 1 << i] = table[b][s] + x; }
        }
      }
    }
  }

  // bitset の [l,r) 部分
  template <typename SUM_TYPE>
  SUM_TYPE query(My_Bitset &x, int l, int r) {
    SUM_TYPE ANS = 0;
    while (l < r && (l & 63)) {
      if (x[l]) ANS += base[l];
      l++;
    }
    while (l < r && (r & 63)) {
      --r;
      if (x[r]) ANS += base[r];
    }
    if (l == r) return ANS;
    l /= 64, r /= 64;
    FOR(b, l, r) {
      u64 s = x.dat[b];
      ANS += table[b * 6 + 0][s >> 0 & 2047];
      ANS += table[b * 6 + 1][s >> 11 & 2047];
      ANS += table[b * 6 + 2][s >> 22 & 2047];
      ANS += table[b * 6 + 3][s >> 33 & 2047];
      ANS += table[b * 6 + 4][s >> 44 & 2047];
      ANS += table[b * 6 + 5][s >> 55 & 2047];
    }
    return ANS;
  }
};
template <typename T, int MAXSIZE>
T Sum_Over_Bit_Positions<T, MAXSIZE>::table[MAXSIZE / 64 * 6 + 10][1 << 11];
#line 12 "main.cpp"

constexpr int th = 1000;
using BS = My_Bitset;

void solve() {
  LL(N);
  STR(S);
  Suffix_Array X(S);

  ll ANS = 0;
  Suffix_Tree<string> ST(S, X);
  auto [G, dat] = ST.build();

  // すべてが good とした場合の ANS
  for (auto &[a, b, c, d]: dat) { ANS += ll(c + d - 1) * (d - c) / 2; }

  FastSet yet(N);
  FOR(i, N) yet.insert(i);

  auto lpf = lpf_table(1 << 18);
  vc<BS> SMALL_P(th);
  FOR(p, th) {
    if (lpf[p] == p) {
      BS x(N + 1);
      for (int i = 0; i <= N; i += p) x[i] = 1;
      SMALL_P[p] = x;
    }
  }

  auto get = [&](int a) -> BS {
    yet.erase(a);
    int n = X.SA[a] + 1; // index
    // あとで
    BS x(N + 1);
    for (auto &[p, e]: factor_by_lpf(n, lpf)) {
      if (p < th)
        x |= SMALL_P[p];
      else {
        for (int i = 0; i <= N; i += p) x[i] = 1;
      }
    }
    return x;
  };

  Sum_Over_Bit_Positions<int, 100'101> SB(N + 1, [&](int i) -> int { return 1 + i; });

  Tree<decltype(G)> tree(G);
  auto dfs = [&](auto &dfs, int v) -> BS {
    assert(tree.head[v] == v);
    auto path = tree.heavy_path_at(v);
    reverse(all(path));

    BS x(N + 1, 1);
    FOR(i, len(path)) {
      int v = path[i];
      for (auto &c: tree.collect_light(v)) { x &= dfs(dfs, c); }
      auto [a, b, c, d] = dat[v];
      if (c == 0) break;
      yet.enumerate(a, b, [&](int i) -> void {
        x &= get(i);
        yet.erase(i);
      });
      --c, --d;
      // FOR(i, c, d) if (x[i]) ANS -= 1 + i;
      ANS -= SB.query<ll>(x, c, d);
    }
    return x;
  };
  dfs(dfs, 0);
  print(ANS);
}

signed main() {
  solve();
  return 0;
}