//注意点
//Tは3つの値を持つ構造
//だがワイルドカードとしても使っている
//#pragma GCC optimize ("-O3")
#include <bits/stdc++.h>
using namespace std;
//@起動時
struct initon {
    initon() {
        cin.tie(0);
        ios::sync_with_stdio(false);
        cout.setf(ios::fixed);
        cout.precision(16);
        srand((unsigned) clock() + (unsigned) time(NULL));
    };
} __initon;
//衝突対策
#define ws ___ws

//@必須構造
struct T {
    int f, s, t;
    T() { f = -1, s = -1, t = -1; }
    T(int f, int s, int t) : f(f), s(s), t(t) {}
    bool operator<(const T &r) const {
        return f != r.f ? f < r.f : s != r.s ? s < r.s : t < r.t;
        //return f != r.f ? f > r.f : s != r.s ? s > r.s : t > r.t; 大きい順
    }
    bool operator>(const T &r) const {
        return f != r.f ? f > r.f : s != r.s ? s > r.s : t > r.t;
        //return f != r.f ? f > r.f : s != r.s ? s > r.s : t > r.t; 小さい順
    }
    bool operator==(const T &r) const {
        return f == r.f && s == r.s && t == r.t;
    }
    bool operator!=(const T &r) const {
        return f != r.f || s != r.s || t != r.t;
    }
    int operator[](int i) {
        assert(i < 3);
        return i == 0 ? f : i == 1 ? s : t;
    }
};
struct F {
    int a, b, c, d;
    F() { a = -1, b = -1, c = -1, d = -1; }
    F(int a, int b, int c, int d) : a(a), b(b), c(c), d(d) {}
    bool operator<(const F &r) const {
        return a != r.a ? a < r.a : b != r.b ? b < r.b : c != r.c ? c < r.c : d < r.d;
//        return a != r.a ? a > r.a : b != r.b ? b > r.b : c != r.c ? c > r.c : d > r.d;
    }
    bool operator>(const F &r) const {
        return a != r.a ? a > r.a : b != r.b ? b > r.b : c != r.c ? c > r.c : d > r.d;
//        return a != r.a ? a < r.a : b != r.b ? b < r.b : c != r.c ? c < r.c : d < r.d;
    }
    bool operator==(const F &r) const {
        return a == r.a && b == r.b && c == r.c && d == r.d;
    }
    bool operator!=(const F &r) const {
        return a != r.a || b != r.b || c != r.c || d != r.d;
    }
    int operator[](int i) {
        assert(i < 4);
        return i == 0 ? a : i == 1 ? b : i == 2 ? c : d;
    }
};

T mt(int a, int b, int c) {
    return T(a, b, c);
}
//@マクロ省略系 型,構造

#define int long long
#define ll long long
#define double long double
#define ull unsigned long long
using dou = double;
using itn = int;
using str = string;
using bo= bool;
using P = pair<ll, ll>;

#define fi first
#define se second
#define vec vector
#define beg begin
#define rbeg rbegin
#define con continue
#define bre break
#define brk break
#define is ==


//マクロ省略系 コンテナ

using vi = vector<int>;
#define _overloadvvi(_1, _2, _3, _4, name, ...) name
#define vvi1(a) vec<vi> a
#define vvi2(a, b) vec<vi> a(b)
#define vvi3(a, b, c) vec<vi> a(b,vi(c))
#define vvi4(a, b, c, d) vec<vi> a(b,vi(c,d))
#define vvi(...) _overloadvvi(__VA_ARGS__,vvi4,vvi3,vvi2 ,vvi1)(__VA_ARGS__)


using vl = vector<ll>;
#define _overloadvvl(_1, _2, _3, _4, name, ...) name
#define vvl1(a) vec<vl> a
#define vvl2(a, b) vec<vl> a(b)
#define vvl3(a, b, c) vec<vl> a(b,vl(c))
#define vvl4(a, b, c, d) vec<vl> a(b,vl(c,d))
#define vvl(...) _overloadvvl(__VA_ARGS__,vvl4,vvl3,vvl2 ,vvl1)(__VA_ARGS__)

using vb = vector<bool>;
#define _overloadvvb(_1, _2, _3, _4, name, ...) name
#define vvb1(a) vec<vb> a
#define vvb2(a, b) vec<vb> a(b)
#define vvb3(a, b, c) vec<vb> a(b,vb(c))
#define vvb4(a, b, c, d) vec<vb> a(b,vb(c,d))
#define vvb(...) _overloadvvb(__VA_ARGS__,vvb4,vvb3,vvb2 ,vvb1)(__VA_ARGS__)

using vs = vector<string>;
#define _overloadvvs(_1, _2, _3, _4, name, ...) name
#define vvs1(a) vec<vs> a
#define vvs2(a, b) vec<vs> a(b)
#define vvs3(a, b, c) vec<vs> a(b,vs(c))
#define vvs4(a, b, c, d) vec<vs> a(b,vs(c,d))
#define vvs(...) _overloadvvs(__VA_ARGS__,vvs4,vvs3,vvs2 ,vvs1)(__VA_ARGS__)

using vd = vector<double>;
#define _overloadvvd(_1, _2, _3, _4, name, ...) name
#define vvd1(a) vec<vd> a
#define vvd2(a, b) vec<vd> a(b)
#define vvd3(a, b, c) vec<vd> a(b,vd(c))
#define vvd4(a, b, c, d) vec<vd> a(b,vd(c,d))
#define vvd(...) _overloadvvd(__VA_ARGS__,vvd4,vvd3,vvd2 ,vvd1)(__VA_ARGS__)

using vc=vector<char>;
#define _overloadvvc(_1, _2, _3, _4, name, ...) name
#define vvc1(a) vec<vc> a
#define vvc2(a, b) vec<vc> a(b)
#define vvc3(a, b, c) vec<vc> a(b,vc(c))
#define vvc4(a, b, c, d) vec<vc> a(b,vc(c,d))
#define vvc(...) _overloadvvc(__VA_ARGS__,vvc4,vvc3,vvc2 ,vvc1)(__VA_ARGS__)

using vp = vector<P>;
#define _overloadvvp(_1, _2, _3, _4, name, ...) name
#define vvp1(a) vec<vp> a
#define vvp2(a, b) vec<vp> a(b)
#define vvp3(a, b, c) vec<vp> a(b,vp(c))
#define vvp4(a, b, c, d) vec<vp> a(b,vp(c,d))

using vt = vector<T>;
#define _overloadvvt(_1, _2, _3, _4, name, ...) name
#define vvt1(a) vec<vt> a
#define vvt2(a, b) vec<vt> a(b)
#define vvt3(a, b, c) vec<vt> a(b,vt(c))
#define vvt4(a, b, c, d) vec<vt> a(b,vt(c,d))

#define v3i(a, b, c, d) vector<vector<vi>> a(b, vector<vi>(c, vi(d)))
#define v3d(a, b, c, d) vector<vector<vd>> a(b, vector<vd>(c, vd(d)))
#define v3m(a, b, c, d) vector<vector<vm>> a(b, vector<vm>(c, vm(d)))

#define _vvi vector<vi>
#define _vvl vector<vl>
#define _vvb vector<vb>
#define _vvs vector<vs>
#define _vvd vector<vd>
#define _vvc vector<vc>
#define _vvp vector<vp>

#define PQ priority_queue<ll, vector<ll>, greater<ll> >
#define tos to_string
using mapi = map<int, int>;
using mapd = map<dou, int>;
using mapc = map<char, int>;
using maps = map<str, int>;
using seti = set<int>;
using setd = set<dou>;
using setc = set<char>;
using sets = set<str>;
using qui = queue<int>;
#define bset bitset
#define uset unordered_set
#define mset multiset
#define umap unordered_map
#define mmap multimap

//マクロ 繰り返し
#define _overloadrep(_1, _2, _3, _4, name, ...) name
# define _rep(i, n) for(int i = 0,_lim=n; i < _lim ; i++)
#define repi(i, m, n) for(int i = m,_lim=n; i < _lim ; i++)
#define repadd(i, m, n, ad) for(int i = m,_lim=n; i < _lim ; i+= ad)
#define rep(...) _overloadrep(__VA_ARGS__,repadd,repi,_rep,)(__VA_ARGS__)
#define _rer(i, n) for(int i = n; i >= 0 ; i--)
#define reri(i, m, n) for(int i = m,_lim=n; i >= _lim ; i--)
#define rerdec(i, m, n, dec) for(int i = m,_lim=n; i >= _lim ; i-=dec)
#define rer(...) _overloadrep(__VA_ARGS__,rerdec,reri,_rer,)(__VA_ARGS__)
#define fora(a, b) for(auto&& a : b)
#define forg(gi, ve) for (int gi = 0, f, t, c; gi < ve.size() && (f = ve[gi].from, t = ve[gi].to, c = ve[gi].cost, true); gi++)
#define fort(gi, ve) for (int gi = 0, f, t, c;gi<ve.size()&& (gi+= (ve[gi].to==p))< ve.size() && (f = ve[gi].from,t=ve[gi].to, c = ve[gi].cost,true); gi++)

//マクロ 定数
#define k3 1010
#define k4 10101
#define k5 101010
#define k6 1010101
#define k7 10101010
const int inf = (int) 1e9 + 100;
const ll linf = (ll) 1e18 + 100;
const double eps = 1e-9;
const double PI = 3.1415926535897932384626433832795029L;
const int y4[] = {-1, 1, 0, 0};
const int x4[] = {0, 0, -1, 1};
const int y8[] = {0, 1, 0, -1, -1, 1, 1, -1};
const int x8[] = {1, 0, -1, 0, 1, -1, 1, -1};

//マクロ省略形 関数等
#define arsz(a) (sizeof(a)/sizeof(a[0]))
#define sz(a) ((int)(a).size())
#define rs resize
#define mp make_pair
#define pb push_back
#define pf push_front
#define eb emplace_back
#define all(a) (a).begin(),(a).end()
#define rall(a) (a).rbegin(),(a).rend()

//@拡張系 こう出来るべきというもの

//埋め込み 存在を意識せずに機能を増やされているもの
// 境界チェック付きvector
namespace std_vector_bounds_checking {
    using namespace std;
    template<class T, class A = std::allocator<T>> struct vector : std::vector<T, A> {
        using std::vector<T, A>::vector;
        typename std::vector<T>::reference operator[](typename std::vector<T>::size_type n) {
            return this->at(n);
        }
    };
}
namespace std {
    template<> class hash<std::pair<signed, signed>> {
    public:
        size_t operator()(const std::pair<signed, signed> &x) const {
            return hash<ll>()(((ll) x.first << 32) + x.second);
        }
    };
    template<> class hash<std::pair<ll, ll>> {
    public:
        size_t operator()(const std::pair<ll, ll> &x) const {
            return hash<ll>()(((ll) x.first << 32) + x.second);
        }
    };
}
istream &operator>>(istream &iss, P &a) {
    iss >> a.first >> a.second;
    return iss;
}
template<typename T> istream &operator>>(istream &iss, vector<T> &vec) {
    for (T &x: vec) iss >> x;
    return iss;
}
ostream &operator<<(ostream &os, P p) {
    os << p.fi << " " << p.se;
    return os;
}
ostream &operator<<(ostream &os, T p) {
    os << p.f << " " << p.s << " " << p.t;
    return os;
}
ostream &operator<<(ostream &os, F p) {
    os << p.a << " " << p.b << " " << p.c << " " << p.d;
    return os;
}
template<typename T> ostream &operator<<(ostream &os, vector <T> &vec) {
    for (int i = 0; i < vec.size(); i++)os << vec[i] << (i + 1 == vec.size() ? "" : " ");
    return os;
}
template<typename T> ostream &operator<<(ostream &os, vector <vector<T>> &vec) {
    for (int i = 0; i < vec.size(); i++) {
        for (int j = 0; j < vec[0].size(); j++) {
            os << vec[i][j];
        }
        os << endl;
    }
    return os;
}
template<typename V, typename H> void resize(vector<V> &vec, const H head) {  //再帰の終端。 可変長templateの長さが 0 になるとこっちが呼ばれる。
    vec.resize(head);
}
template<typename V, typename H, typename ... T> void resize(vector<V> &vec, const H &head, const T ... tail) {
    vec.resize(head);
    for (auto &v: vec) resize(v, tail...);
}
template<class T> T pop(set<T> &set) {
    T res = *set.begin();
    set.erase(set.find(res));
    return res;
}
template<class T> T pop(mset<T> &set) {
    T res = *set.begin();
    set.erase(set.find(res));
    return res;
}
template<class T> T popBack(set<T> &set) {
    T res = *set.rbegin();
    set.erase(set.find(res));
    return res;
}
template<class T> T popBack(mset<T> &set) {
    T res = *set.rbegin();
    set.erase(set.find(res));
    return res;
}
inline void sort(string &a) { sort(a.begin(), a.end()); }
template<class T> inline void sort(vector<T> &a) { sort(a.begin(), a.end()); };
template<class T> inline void sort(vector<T> &a, int len) { sort(a.begin(), a.begin() + len); };
template<class T, class F> inline void sort(vector<T> &a, F f) { sort(a.begin(), a.end(), [&](T l, T r) { return f(l) < f(r); }); };
template<class T> inline void rsort(vector<T> &a) { sort(a.begin(), a.end(), greater<T>()); };
template<class T> inline void rsort(vector<T> &a, int len) { sort(a.begin(), a.begin() + len, greater<T>()); };
template<class U, class F> inline void rsort(vector<U> &a, F f) { sort(a.begin(), a.end(), [&](U l, U r) { return f(l) > f(r); }); };
template<class U> inline void sortp(vector<U> &a, vector<U> &b) {
    vp c;
    int n = sz(a);
    assert(n == sz(b));
    rep(i, n)c.eb(a[i], b[i]);
    sort(c);
    rep(i, n) {
        a[i] = c[i].first;
        b[i] = c[i].second;;
    }
};
template<class U, class F> inline void sortp(vector<U> &a, vector<U> &b, F f) {
    vp c;
    int n = sz(a);
    assert(n == sz(b));
    rep(i, n)c.eb(a[i], b[i]);
    sort(c, f);
    rep(i, n) {
        a[i] = c[i].first;
        b[i] = c[i].second;
    }
};
template<class U, class F> inline void rsortp(vector<U> &a, vector<U> &b) {
    vp c;
    int n = sz(a);
    assert(n == sz(b));
    rep(i, n)c.eb(a[i], b[i]);
    rsort(c);
    rep(i, n) {
        a[i] = c[i].first;
        b[i] = c[i].second;
    }
};
template<class U, class F> inline void rsortp(vector<U> &a, vector<U> &b, F f) {
    vp c;
    int n = sz(a);
    assert(n == sz(b));
    rep(i, n)c.eb(a[i], b[i]);
    rsort(c, f);
    rep(i, n) {
        a[i] = c[i].first;
        b[i] = c[i].second;
    }
};
template<class U> inline void sortt(vector<U> &a, vector<U> &b, vector<U> &c) {
    vt r;
    int n = sz(a);
    assert(n == sz(b));
    assert(n == sz(c));
    rep(i, n)r.eb(a[i], b[i], c[i]);
    sort(r);
    rep(i, n) {
        a[i] = r[i].f;
        b[i] = r[i].s;
        c[i] = r[i].t;
    }
};
template<class U, class F> inline void sortt(vector<U> &a, vector<U> &b, vector<U> &c, F f) {
    vt r;
    int n = sz(a);
    assert(n == sz(b));
    assert(n == sz(c));
    rep(i, n)r.eb(a[i], b[i], c[i]);
    sort(r, f);
    rep(i, n) {
        a[i] = r[i].f;
        b[i] = r[i].s;
        c[i] = r[i].t;
    }
};
template<class U, class F> inline void rsortt(vector<U> &a, vector<U> &b, vector<U> &c, F f) {
    vt r;
    int n = sz(a);
    assert(n == sz(b));
    assert(n == sz(c));
    rep(i, n)r.eb(a[i], b[i], c[i]);
    rsort(r, f);
    rep(i, n) {
        a[i] = r[i].f;
        b[i] = r[i].s;
        c[i] = r[i].t;
    }
};
template<class T> inline void sort2(vector<vector<T>> &a) { for (int i = 0, n = a.size(); i < n; i++)sort(a[i]); }
template<class T> inline void rsort2(vector<vector<T>> &a) { for (int i = 0, n = a.size(); i < n; i++)rsort(a[i]); }
template<typename A, size_t N, typename T> void fill(A (&a)[N], const T &v) { rep(i, N)a[i] = v; }
template<typename A, size_t N, size_t O, typename T> void fill(A (&a)[N][O], const T &v) { rep(i, N)rep(j, O)a[i][j] = v; }
template<typename A, size_t N, size_t O, size_t P, typename T> void fill(A (&a)[N][O][P], const T &v) { rep(i, N)rep(j, O)rep(k, P)a[i][j][k] = v; }
template<typename A, size_t N, size_t O, size_t P, size_t Q, typename T> void fill(A (&a)[N][O][P][Q], const T &v) { rep(i, N)rep(j, O)rep(k, P)rep(l, Q)a[i][j][k][l] = v; }
template<typename A, size_t N, size_t O, size_t P, size_t Q, size_t R, typename T> void fill(A (&a)[N][O][P][Q][R], const T &v) { rep(i, N)rep(j, O)rep(k, P)rep(l, Q)rep(m, R)a[i][j][k][l][m] = v; }
template<typename A, size_t N, size_t O, size_t P, size_t Q, size_t R, size_t S, typename T> void fill(A (&a)[N][O][P][Q][R][S], const T &v) { rep(i, N)rep(j, O)rep(k, P)rep(l, Q)rep(m, R)rep(n, S)a[i][j][k][l][m][n] = v; }

template<typename V, typename T>
void fill(V &xx, const T vall) {
    xx = vall;
}
template<typename V, typename T>
void fill(vector<V> &vecc, const T vall) {
    for (auto &&vx: vecc) fill(vx, vall);
}

//@汎用便利関数 入力
template<typename T = int> T in() {
    T x;
    cin >> x;
    return (x);
}
string sin() { return in<string>(); }
double din() { return in<double>(); }
ll lin() { return in<ll>(); }
#define na(a, n) a.resize(n); rep(i,n) cin >> a[i];
#define nao(a, n) a.resize(n+1); rep(i,n) cin >> a[i+1];
#define nad(a, n) a.resize(n); rep(i,n) cin >> a[i], a[i]--;
#define na2(a, b, n) a.resize(n),b.resize(n);rep(i, n)cin >> a[i] >> b[i];
#define na2d(a, b, n) a.resize(n),b.resize(n);rep(i, n)cin >> a[i] >> b[i],a[i]--,b[i]--;
#define na3(a, b, c, n) a.resize(n),b.resize(n),c.resize(n);   rep(i, n)cin >> a[i] >> b[i] >> c[i];
#define na3d(a, b, c, n) a.resize(n),b.resize(n),c.resize(n);   rep(i, n)cin >> a[i] >> b[i] >> c[i],a[i]--,b[i]--,c[i]--;
#define nt(a, h, w) resize(a,h,w);rep(hi,h)rep(wi,w) cin >> a[hi][wi];
#define ntd(a, h, w) rs(a,h,w);rep(hi,h)rep(wi,w) cin >> a[hi][wi], a[hi][wi]--;
#define ntp(a, h, w) fill(a,'#');rep(hi,1,h+1)rep(wi,1,w+1) cin >> a[hi][wi];

//汎用便利関数 出力
template<class T> void out(T x) { typeid(x) == typeid(double) ? cout << fixed << setprecision(10) << x << endl : cout << x << endl; }
template<class T> void out(vector<T> a) {
    fora(v, a)cout << v << endl;
}
//デバッグ
#define sp << " " <<

#define debugName(VariableName) # VariableName

#define _deb1(x) cerr <<  debugName(x)<<" = "<<x << endl
#define _deb2(x, y) cerr <<  debugName(x)<<" = "<<x<<", "<< debugName(y)<<" = "<<y<< endl
#define _deb3(x, y, z) cerr <<  debugName(x)<<" = "<<x  << ", " <<  debugName(y)<<" = "<<y <<", " debugName(z)<<" = "<<z <<endl
#define _deb4(x, y, z, a) cerr <<  debugName(x)<<" = "<<x <<", " <<   debugName(y)<<" = "<<y <<", " <<  debugName(z)<<" = "<<z <<", " <<  debugName(a)<<" = "<<a<<endl
#define _deb5(x, y, z, a, b) cerr <<  debugName(x)<<" = "<<x <<", " <<   debugName(y)<<" = "<<y <<", " <<  debugName(z)<<" = "<<z <<", " <<  debugName(a)<<" = "<<a<<", " <<  debugName(b)<<" = "<<b<<endl


#define _overloadebug(_1, _2, _3, _4, _5, name, ...) name
#define debug(...) _overloadebug(__VA_ARGS__,_deb5,_deb4,_deb3,_deb2,_deb1)(__VA_ARGS__)
#define deb(...) _overloadebug(__VA_ARGS__,_deb5,_deb4,_deb3,_deb2,_deb1)(__VA_ARGS__)


#define debugline(x) cerr << x << " " << "(L:" << __LINE__ << ")" << '\n'

//よく使うクラス、構造体
class UnionFind {
public:
    vi par, rank, sizes;
    int n, trees;
    UnionFind(int n) : n(n), trees(n) {
        par.resize(n), rank.resize(n), sizes.resize(n);
        rep(i, n)par[i] = i, sizes[i] = 1;
    }
    int root(int x) {
        if (par[x] == x)return x;
        else return par[x] = root(par[x]);
    }
    int find(int x) { return root(x); }
    void unite(int x, int y) {
        x = root(x);
        y = root(y);
        if (x == y)return;
        if (rank[x] < rank[y])swap(x, y);
        trees--;
        par[y] = x;
        sizes[x] += sizes[y];
        if (rank[x] == rank[y])rank[x]++;
    }
    bool same(int x, int y) { return root(x) == root(y); }
    int size(int x) { return sizes[root(x)]; }
    //順不同 umapなので
    vec<vi> sets() {
        vec<vi> res(trees);
        umap<int, vi> map;
        rep(i, n) map[root(i)].push_back(i);
        int i = 0;
        for (auto &&p:map) {
            int r = p.fi;
            res[i].push_back(r);
            for (auto &&v:p.se) {
                if (r == v)continue;
                res[i].push_back(v);
            }
            i++;
        }
        return res;
    }
};

//MOD関連
ll MOD = (int) 1e9 + 7;
class mint {
public:
    int x;
    mint() : x(0) {}
    mint(signed y) : x(y >= 0 ? y % MOD : MOD - (-y) % MOD) {}
    mint(int y) : x(y >= 0 ? y % MOD : MOD - (-y) % MOD) {}
    static int _mpow(int v, ll a) {
        ll x = v, n = a, res = 1;
        while (n) {
            if (n & 1)res = (res * x) % MOD;
            x = (x * x) % MOD;
            n >>= 1;
        }
        return res;
    }

    // Arithmetic Oprators
    mint &operator+=(mint that) {
        if ((x += that.x) >= MOD) x -= MOD;
        return *this;
    }
    mint &operator-=(mint that) {
        if ((x += MOD - that.x) >= MOD) x -= MOD;
        return *this;
    }
    mint &operator*=(mint that) {
        x = 1LL * x * that.x % MOD;
        return *this;
    }
    mint &operator/=(const mint &that);
    mint &operator^=(const mint &that) {
        this->x = _mpow(x, that.x);
        return *this;
    }
    mint &operator%=(mint that) {
        x %= that.x;
        return *this;
    }

    mint &operator+=(const int that) { return *this += mint(that); }
    mint &operator-=(const int that) { return *this -= mint(that); }
    mint &operator*=(const int that) { return *this *= mint(that); }
    mint &operator/=(const int that);
    mint &operator^=(const int that) {
        this->x = _mpow(x, that);
        return *this;
    }
    mint &operator%=(const int that) { return *this %= mint(that); }

    mint &operator+=(const signed that) { return *this += mint(that); }
    mint &operator-=(const signed that) { return *this -= mint(that); }
    mint &operator*=(const signed that) { return *this *= mint(that); }
    mint &operator/=(const signed that);
    mint &operator^=(const signed that) {
        this->x = _mpow(x, that);
        return *this;
    }
    mint &operator%=(const signed that) { return *this %= mint(that); }

    // Comparators
    bool operator<(mint that) { return x < that.x; }
    bool operator>(mint that) { return x > that.x; }
    bool operator<=(mint that) { return x <= that.x; }
    bool operator>=(mint that) { return x >= that.x; }
    bool operator!=(mint that) { return x != that.x; }
    bool operator==(mint that) { return x == that.x; }
    bool operator!=(int that) { return x != that; }
    bool operator==(int that) { return x == that; }
    bool operator!=(signed that) { return x != that; }
    bool operator==(signed that) { return x == that; }

    // Utilities
    unsigned getval() const { return x; }
    operator int() { return x; }
    mint operator+(mint that) const { return mint(*this) += that; }
    mint operator-(mint that) const { return mint(*this) -= that; }
    mint operator*(mint that) const { return mint(*this) *= that; }
    mint operator%(mint that) const { return mint(*this) %= that; }
    mint operator+(const int that) const { return mint(*this) += that; }
    mint operator-(const int that) const { return mint(*this) -= that; }
    mint operator*(const int that) const { return mint(*this) *= that; }
    mint operator%(const int that) const { return mint(*this) %= that; }
    mint operator=(const int that) { return *this = mint(that); }
    mint operator+(const signed that) const { return mint(*this) += that; }
    mint operator-(const signed that) const { return mint(*this) -= that; }
    mint operator*(const signed that) const { return mint(*this) *= that; }
    mint operator%(const signed that) const { return mint(*this) %= that; }
    mint operator=(const signed that) { return *this = mint(that); }
    mint operator++() {
        x++;
        return *this;
    }
    mint operator++(signed) {
        auto ret = *this;
        x++;
        return ret;
    }
    friend void operator+=(ll &a, const mint &b) { a = mint(a % MOD + b.x); }
    friend void operator-=(ll &a, const mint &b) { a = mint(a % MOD - b.x); }
    friend void operator*=(ll &a, const mint &b) { a = mint(a % MOD * b.x); }
    friend void operator/=(ll &a, const mint &b);
    friend mint operator+(const ll a, const mint &b) { return mint(a % MOD + b.x); }
    friend mint operator-(const ll a, const mint &b) { return mint(a % MOD - b.x); }
    friend mint operator*(const ll a, const mint &b) { return mint(a % MOD * b.x); }
    friend mint operator^(const ll a, const mint &b) { return _mpow(a, b.x); }
};
mint itom(int v) {
    mint res;
    res.x = v;
    return res;
}
const int setModMax = 510000;
vector<mint> fac, finv, inv;
void setMod(int m = MOD) {
    fac.resize(setModMax);
    finv.resize(setModMax);
    inv.resize(setModMax);
    MOD = m;
    fac[0] = fac[1] = 1;
    finv[0] = finv[1] = 1;
    inv[1] = 1;
    for (int i = 2; i < setModMax; i++) {
        fac[i] = fac[i - 1].x * i % MOD;
        inv[i] = MOD - inv[MOD % i].x * (MOD / i) % MOD;
        finv[i] = finv[i - 1].x * inv[i].x % MOD;
    }
}
mint mpow(int v, ll a) {
    return mint::_mpow(v, a);
}
mint com(ll n, ll r) {
    if (n < r || n < 0 || r < 0)return 0;
    if (fac.size() == 0)setMod();
    return fac[n] * finv[r] * finv[n - r];
}
mint ncr(ll n, ll r) { return com(n, r); }
mint nhr(ll n, ll r) { return com(n + r - 1, r); }

//拡張ユークリッドの互除法
mint minv(ll a) {
    if (fac[0] == 0)setMod();
    if (a < setModMax) return inv[a];
    a %= MOD;
    ll b = MOD, x = 1, y = 0;
    while (b) {
        ll t = a / b;
        a -= t * b;
        swap(a, b);
        x -= t * y;
        swap(x, y);
    }
    return x;
}
mint &mint::operator/=(const mint &that) { return *this *= minv(that.x); }
mint &mint::operator/=(const ll a) { return *this *= minv(a); }
mint &mint::operator/=(const signed a) { return *this *= minv(a); }
void operator/=(ll &a, const mint &b) { a = (a * minv(b.x)).x; }
using vm=vector<mint>;
#define vvm(a, b, c) vec<vm> a(b,vm(c))
using PM = pair<mint, mint>;
vb isPrime;
vi primes;

void setPrime() {
    isPrime.resize(4010101);
    fill(isPrime, true);
    int len = sizeof(isPrime) / sizeof(isPrime[0]);
    isPrime[0] = isPrime[1] = false;
    for (int i = 2; i <= sqrt(len) + 5; ++i) {
        if (!isPrime[i])continue;
        for (int j = 2; i * j < len; ++j) {
            isPrime[i * j] = false;
        }
    }
    rep(i, len)if (isPrime[i])primes.pb(i);
}
//幾何 Pをcomplexとして扱う
bool eq(double a, double b) { return fabs(a - b) < eps; }

using C =complex<double>;

C rot(C &a, dou th) {
    return a * C(cos(th), sin(th));
}

dou inpro(C &a, C &b) {
    return real(a * conj(b));
}

//90度回転させて内積が0なら平行
bool line(C a, C b, C c) {
    C ab = b - a;
    C ac = c - a;
    //複素数の掛け算は回転
    ab *= C(0, 1);
    return eq(inpro(ab, ac), 0);
}
bool line(P a, P b, P c) {
    return line(C(a.fi, a.se), C(b.fi, b.se), C(c.fi, c.se));
}

bool line(int xa, int ya, int xb, int yb, int xc, int yc) {
    C a = C(xa, ya);
    C b = C(xb, yb);
    C c = C(xc, yc);
    return line(a, b, c);
}
//便利関数

//テスト用
char ranc() {
    return (char) ('a' + rand() % 26);
}
int rand(int min, int max) {
    assert(min <= max);
    if (min >= 0 && max >= 0) {
        return rand() % (max + 1 - min) + min;
    } else if (max < 0) {
        return -rand(-max, -min);
    } else {
        //+
        if (rand() % 2) {
            return rand(0, max);
            //-
        } else {
            return -rand(0, -min);
        }
    }
}
vi ranv(int n, int min, int max) {
    vi v(n);
    rep(i, n)v[i] = rand(min, max);
    return v;
}
//単調増加
vi ranvi(int n, int min, int max) {
    vi v(n);
    bool bad = 1;
    while (bad) {
        bad = 0;
        v.resize(n);
        rep(i, n) {
            if (i && min > max - v[i - 1]) {
                bad = 1;
                break;
            }
            if (i)v[i] = v[i - 1] + rand(min, max - v[i - 1]);
            else v[i] = rand(min, max);
        }
    }
    return v;
}
void ranvlr(int n, int min, int max, vi &l, vi &r) {
    l.resize(n);
    r.resize(n);
    rep(i, n) {
        l[i] = rand(min, max);
        r[i] = l[i] + rand(0, max - l[i]);
    }
}
//便利 汎用
template<typename V, typename T> int find(vector<V> &a, const T key) {
    rep(i, sz(a))if (a[i] == key)return i;
    return -1;
}
template<typename V, typename T> P find(vector<vector<V>> &a, const T key) {
    rep(i, sz(a)) rep(j, sz(a[0]))if (a[i][j] == key)return mp(i, j);
    return mp(-1, -1);
}
template<typename V, typename U> T find(vector<vector<vector<V>>> &a, const U key) {
    rep(i, sz(a))rep(j, sz(a[0]))rep(k, sz(a[0][0]))if (a[i][j][k] == key)return mt(i, j, k);
    return mt(-1, -1, -1);
}

template<typename V, typename T> int count(V &a, const T k) {
    return a == k;
}
template<typename V, typename T> int count(vector<V> &a, const T k) {
    int ret = 0;
    fora(v, a)ret += count(v, k);
    return ret;
}

int mod(int a, int m) {
    return (a % m + m) % m;
}
int pow(int a) { return a * a; };
void iota(vector<int> &ve, int s, int n) {
    ve.resize(n);
    iota(all(ve), s);
}
vi iota(int s, int n) {
    vi ve;
    ve.resize(n);
    iota(all(ve), s);
    return ve;
}
void ole() {
#ifdef _DEBUG
    debugline("ole");
    exit(0);
#endif
    string a = "a";
    rep(i, 30)a += a;
    rep(i, 1 << 17)cout << a << endl;
    cout << "OLE 出力長制限超過" << endl;
    exit(0);
}
void tle() { while (inf)cout << inf << endl; }
ll gcd(ll a, ll b) { return b ? gcd(b, a % b) : a; }
ll gcd(vi b) {
    ll res = b[0];
    for (auto &&v :b)res = gcd(v, res);
    return res;
}
ll lcm(ll a, ll b) { return a / gcd(a, b) * b; }
ll rev(ll a) {
    ll res = 0;
    while (a) {
        res *= 10;
        res += a % 10;
        a /= 10;
    }
    return res;
}
template<class T> vector<T> rev(vector<T> &a) {
    vector<T> ret = a;
    reverse(all(ret));
    return ret;
}
ll ceil(ll a, ll b) {
    if (b == 0) {
        debugline("ceil");
        deb(a, b);
        ole();
        return -1;
    } else return (a + b - 1) / b;
}
ll sqrt(ll a) {
    if (a < 0) {
        debugline("sqrt");
        deb(a);
        ole();
    }
    ll res = (ll) std::sqrt(a);
    while (res * res < a)res++;
    return res;
}
double log(double e, double x) { return log(x) / log(e); }
ll sig(ll t) { return (1 + t) * t / 2; }
ll sig(ll s, ll t) { return (s + t) * (t - s + 1) / 2; }

vi divisors(int v) {
    vi res;
    for (int i = 1; i <= sqrt(v); ++i) {
        if (v % i == 0) {
            res.pb(i);
            if (i != v / i)res.pb(v / i);
        }
    }
    return res;
}

vi factorization(int v) {
    int tv = v;
    vi res;
    if (isPrime.size() == 0)setPrime();
    for (auto &&p :primes) {
        if (v % p == 0)res.push_back(p);
        while (v % p == 0) {
            v /= p;
        }
        if (v == 1 || p * p > tv)break;
    }
    if (v > 1)res.pb(v);
    return res;
}

unordered_map<int, int> factorizationMap(int v) {
    int tv = v;
    unordered_map<int, int> res;
    if (isPrime.size() == 0)setPrime();
    for (auto &&p :primes) {
        while (v % p == 0) {
            res[p]++;
            v /= p;
        }
        if (v == 1 || p * p > tv)break;
    }
    if (v > 1)res[v]++;
    return res;
}
int get(int a, int keta) { return (a / (int) pow(10, keta)) % 10; }
int keta(int v) {
    int cou = 0;
    while (v) { cou++, v %= 10; }
    return cou;
}

template<class T> void imo(vector<T> &v) {
    int n = v.size();
    rep(i, n - 1)v[i + 1] += v[i];
}

//変換系
template<class T, class U> vector<T> keys(vector<pair<T, U>> a) {
    vector<T> res;
    for (auto &&k :a)res.pb(k.fi);
    return res;
}
template<class T, class U> vector<U> keys(map<T, U> a) {
    vector<U> res;
    for (auto &&k :a)res.pb(k.fi);
    return res;
}

template<class T, class U> vector<U> keys(umap<T, U> a) {
    vector<U> res;
    for (auto &&k :a)res.pb(k.fi);
    return res;
}

template<class T, class U> vector<U> values(vector<pair<T, U>> a) {
    vector<U> res;
    for (auto &&k :a)res.pb(k.se);
    return res;
}

template<class T, class U> vector<T> values(map<T, U> a) {
    vector<T> res;
    for (auto &&k :a)res.pb(k.se);
    return res;
}

template<class T, class U> vector<T> values(umap<T, U> a) {
    vector<T> res;
    for (auto &&k :a)res.pb(k.se);
    return res;
}

vi list(int a) {
    vi res;
    while (a) {
        res.insert(res.begin(), a % 10);
        a /= 10;
    }
    return res;
}

template<class T, class U> bool chmax(T &a, const U &b) {
    if (a < b) {
        a = b;
        return true;
    }
    return false;
}

template<class T, class U> bool chmin(T &a, const U &b) {
    if (b < a) {
        a = b;
        return true;
    }
    return false;
}

template<class T> T min(T a, signed b) {
    return a < b ? a : b;
}

template<class T> T max(T a, signed b) {
    return a < b ? b : a;
}

template<class T> T min(T a, T b, T c) {
    return a >= b ? b >= c ? c : b : a >= c ? c : a;
}
template<class T> T max(T a, T b, T c) {
    return a <= b ? b <= c ? c : b : a <= c ? c : a;
}

template<class T> T min(vector<T> a) {
    return *min_element(all(a));
}
template<class T> T min(vector<T> a, int n) {
    return *min_element(a.begin(), a.begin() + min(n, sz(a)));
}
template<class T> T min(vector<T> a, int s, int n) {
    return *min_element(a.begin() + s, a.begin() + min(n, sz(a)));
}
template<class T> T max(vector<T> a) {
    return *max_element(all(a));
}
template<class T> T max(vector<T> a, int n) {
    return *max_element(a.begin(), a.begin() + min(n, sz(a)));
}
template<class T> T max(vector<T> a, int s, int n) {
    return *max_element(a.begin() + s, a.begin() + min(n, sz(a)));
}
template<typename A, size_t N> A max(A (&a)[N]) {
    A res = a[0];
    rep(i, N)res = max(res, a[i]);
    return res;
}
template<typename A, size_t N, size_t O> A max(A (&a)[N][O]) {
    A res = max(a[0]);
    rep(i, N)res = max(res, max(a[i]));
    return res;
}
template<typename A, size_t N, size_t O, size_t P> A max(A (&a)[N][O][P]) {
    A res = max(a[0]);
    rep(i, N)res = max(res, max(a[i]));
    return res;
}
template<typename A, size_t N, size_t O, size_t P, size_t Q> A max(A (&a)[N][O][P][Q], const T &v) {
    A res = max(a[0]);
    rep(i, N)res = max(res, max(a[i]));
    return res;
}
template<typename A, size_t N, size_t O, size_t P, size_t Q, size_t R> A max(A (&a)[N][O][P][Q][R]) {
    A res = max(a[0]);
    rep(i, N)res = max(res, max(a[i]));
    return res;
}
template<typename A, size_t N, size_t O, size_t P, size_t Q, size_t R, size_t S> A max(A (&a)[N][O][P][Q][R][S]) {
    A res = max(a[0]);
    rep(i, N)res = max(res, max(a[i]));
    return res;
}
template<typename A, size_t N> A min(A (&a)[N]) {
    A res = a[0];
    rep(i, N)res = min(res, a[i]);
    return res;
}
template<typename A, size_t N, size_t O> A min(A (&a)[N][O]) {
    A res = min(a[0]);
    rep(i, N)res = min(res, max(a[i]));
    return res;
}
template<typename A, size_t N, size_t O, size_t P> A min(A (&a)[N][O][P]) {
    A res = min(a[0]);
    rep(i, N)res = min(res, min(a[i]));
    return res;
}
template<typename A, size_t N, size_t O, size_t P, size_t Q> A min(A (&a)[N][O][P][Q], const T &v) {
    A res = min(a[0]);
    rep(i, N)res = min(res, min(a[i]));
    return res;
}
template<typename A, size_t N, size_t O, size_t P, size_t Q, size_t R> A min(A (&a)[N][O][P][Q][R]) {
    A res = min(a[0]);
    rep(i, N)res = min(res, min(a[i]));
    return res;
}
template<typename A, size_t N, size_t O, size_t P, size_t Q, size_t R, size_t S> A min(A (&a)[N][O][P][Q][R][S]) {
    A res = min(a[0]);
    rep(i, N)res = min(res, min(a[i]));
    return res;
}


template<class T> T sum(vector<T> v, int len = -1) {
    if (len == -1)len = v.size();
    T res = 0;
    chmin(len, v.size());
    rep(i, len)res += v[i];
    return res;
}
template<class T> T sum(vector<vector<T>> &v, int h = -1, int w = -1) {
    if (h == -1)h = v.size();
    if (w == -1)w = v[0].size();
    T res = 0;
    chmin(h, v.size());
    chmin(w, v[0].size());
    rep(i, h)rep(j, w)res += v[i][j];
    return res;
}

P sump(vp &v, int len = -1) {
    if (len == -1)len = v.size();
    P res = {0, 0};
    chmin(len, v.size());
    rep(i, len) {
        res.fi += v[i].fi;
        res.se += v[i].se;
    }
    return res;
}

///要素が0の時、返り値は0か1か
template<class T> T mul(vector<T> &v, int len = -1) {
    if (len == -1)len = v.size();
    T res = 1;
    chmin(len, v.size());
    rep(i, len)res *= v[i];
    return res;
}
void clear(PQ &q) { while (q.size())q.pop(); }
template<class T> void clear(queue<T> &q) { while (q.size())q.pop(); }
template<class T> T *negarr(int size) {
    T *body = (T *) malloc((size * 2 + 1) * sizeof(T));
    return body + size;
}
template<class T> T *negarr2(int h, int w) {
    double **dummy1 = new double *[2 * h + 1];
    double *dummy2 = new double[(2 * h + 1) * (2 * w + 1)];
    dummy1[0] = dummy2 + w;
    for (int i = 1; i <= 2 * h + 1; i++) {
        dummy1[i] = dummy1[i - 1] + 2 * w + 1;
    }
    double **a = dummy1 + h;
}
template<class T> struct ruic {
    const vector<T> rui;
    ruic(vector<T> &ru) : rui(ru) {}
    T operator()(int l, int r) {
        assert(l <= r);
        return rui[r] - rui[l];
    }
    T operator[](int i) {
        return rui[i];
    }
};

template<class T> struct rruic {
    const T *rrui;
    rruic(T *ru) : rrui(ru) {}
    //n-1から-1へ
    T operator()(int l, int r) {
        assert(l >= r);
        return rrui[r] - rrui[l];
    }
    T operator[](int i) {
        return rrui[i];
    }
};
//template<class T> vector<T> ruiv(vector<T> &a)
template<class T> ruic<T> ruiv(vector<T> &a) {
    vector<T> res(a.size() + 1);
    rep(i, a.size())res[i + 1] = res[i] + a[i];
    return ruic<T>(res);
}

template<class T> vector<T> ruim(vector<T> &a) {
    vector<T> res(a.size() + 1, 1);
    rep(i, a.size())res[i + 1] = res[i] * a[i];
    return res;
}
//template<class T> T *rrui(vector<T> &a) {
//右から左にかけての半開区間 (-1 n-1]
template<class T> rruic<T> rrui(vector<T> &a) {
    int len = a.size();
    T *body = (T *) malloc((len + 1) * sizeof(T));
    T *res = body + 1;
    rer(i, len - 1)res[i - 1] = res[i] + a[i];
    return rruic<T>(res);
}
//掛け算
template<class T> T *rruim(vector<T> &a) {
    int len = a.size();
    T *body = (T *) malloc((len + 1) * sizeof(T));
    T *res = body + 1;
    res[len - 1] = 1;
    rer(i, len - 1)res[i - 1] = res[i] * a[i];
    return res;
}
template<class T, class U> void inc(T &a, U v = 1) { a += v; }
template<class T, class U> void inc(vector<T> &a, U v = 1) { for (auto &u :a)inc(u, v); }
template<class T> void inc(vector<T> &a) { for (auto &u :a)inc(u, 1); }
template<class T, class U> void dec(T &a, U v = 1) { a -= v; }
template<class T, class U> void dec(vector<T> &a, U v = 1) { for (auto &u :a)dec(u, v); }
template<class T> void dec(vector<T> &a) { for (auto &u :a)dec(u, 1); }
inline bool inside(int h, int w, int H, int W) { return h >= 0 && w >= 0 && h < H && w < W; }
ll u(ll a) { return a < 0 ? 0 : a; }
template<class T> vector<T> u(const vector<T> &a) {
    vector<T> ret = a;
    fora(v, ret)v = u(v);
    return ret;
}
#define MIN(a) numeric_limits<a>::min()
#define MAX(a) numeric_limits<a>::max()
ll goldd(ll left, ll right, function<ll(ll)> calc) {
    double GRATIO = 1.6180339887498948482045868343656;
    ll lm = left + (ll) ((right - left) / (GRATIO + 1.0));
    ll rm = lm + (ll) ((right - lm) / (GRATIO + 1.0));
    ll fl = calc(lm);
    ll fr = calc(rm);
    while (right - left > 10) {
        if (fl < fr) {
            right = rm;
            rm = lm;
            fr = fl;
            lm = left + (ll) ((right - left) / (GRATIO + 1.0));
            fl = calc(lm);
        } else {
            left = lm;
            lm = rm;
            fl = fr;
            rm = lm + (ll) ((right - lm) / (GRATIO + 1.0));
            fr = calc(rm);
        }
    }
    ll minScore = MAX(ll);
    ll resIndex = left;
    for (ll i = left; i < right + 1; i++) {
        ll score = calc(i);
        if (minScore > score) {
            minScore = score;
            resIndex = i;
        }
    }
    return resIndex;
}
ll goldt(ll left, ll right, function<ll(ll)> calc) {
    double GRATIO = 1.6180339887498948482045868343656;
    ll lm = left + (ll) ((right - left) / (GRATIO + 1.0));
    ll rm = lm + (ll) ((right - lm) / (GRATIO + 1.0));
    ll fl = calc(lm);
    ll fr = calc(rm);
    while (right - left > 10) {
        if (fl > fr) {
            right = rm;
            rm = lm;
            fr = fl;
            lm = left + (ll) ((right - left) / (GRATIO + 1.0));
            fl = calc(lm);
        } else {
            left = lm;
            lm = rm;
            fl = fr;
            rm = lm + (ll) ((right - lm) / (GRATIO + 1.0));
            fr = calc(rm);
        }
    }
    if (left > right) {
        ll l = left;
        left = right;
        right = l;
    }
    ll maxScore = MIN(ll);
    ll resIndex = left;
    for (ll i = left; i < right + 1; i++) {
        ll score = calc(i);
        if (maxScore < score) {
            maxScore = score;
            resIndex = i;
        }
    }
    return resIndex;
}
template<class T> T min(vector<vector<T>> &a) {
    T res = MAX(T);
    rep(i, a.size())chmin(res, *min_element(all(a[i])));
    return res;
}
template<class T> T max(vector<vector<T>> &a) {
    T res = MIN(T);
    rep(i, a.size())chmax(res, *max_element(all(a[i])));
    return res;
}

bool bget(ll m, int keta) { return (m >> keta) & 1; }
int bget(ll m, int keta, int sinsuu) {
    m /= (ll) pow(sinsuu, keta);
    return m % sinsuu;
}
inline ll bit(int n) { return (1LL << (n)); }
inline ll bit(int n, int sinsuu) { return (ll) pow(sinsuu, n); }
int bcou(ll m) { return __builtin_popcount(m & 0xFFFFFFFF) + __builtin_popcount(m >> 32); }
//初期化は0を渡す
ll nextComb(ll &mask, int n, int r) {
    if (!mask)return mask = (1LL << r) - 1;
    ll x = mask & -mask; //最下位の1
    ll y = mask + x; //連続した下の1を繰り上がらせる
    ll res = ((mask & ~y) / x >> 1) | y;
    if (bget(res, n))return mask = 0;
    else return mask = res;
}

//n桁以下でビットがr個立っているもののvectorを返す
vl bitCombList(int n, int r) {
    vl res;
    int m = 0;
    while (nextComb(m, n, r)) {
        res.pb(m);
    }
    return res;
}
//大文字小文字を区別する
int altoiaZ(char c) {
    if ('A' <= c && c <= 'Z')return c - 'A';
    return c - 'a' + 26;
}
char itoalaZ(int i) {
    if (i < 26)return 'A' + i;
    return 'a' + i - 26;
}
//aもAも0を返す 基本小文字
int altoi(char c) {
    if ('A' <= c && c <= 'Z')return c - 'A';
    return c - 'a';
}
char itoal(int i) {
    return 'a' + i;
}
int ctoi(char c) { return c - '0'; }
char itoc(int i) { return i + '0'; }
int vtoi(vi &v) {
    int res = 0;
    if (sz(v) > 18) {
        debugline("vtoi");
        deb(sz(v));
        ole();
    }
    rep(i, sz(v)) {
        res *= 10;
        res += v[i];
    }
    return res;
}
vi itov(int i) {
    vi res;
    while (i) {
        res.pb(i % 10);
        i /= 10;
    }
    rev(res);
    return res;
}
vector<vector<int>> ctoi(vector<vector<char>> s, char c) {
    int n = sz(s), m = sz(s[0]);
    vector<vector<int>> res(n, vector<int>(m));
    rep(i, n)rep(j, m)res[i][j] = s[i][j] == c;
    return res;
}
#define UNIQUE(v) v.erase( unique(v.begin(), v.end()), v.end() );

void compress(vi &a) {
    vi b;
    int len = a.size();
    for (int i = 0; i < len; ++i) {
        b.push_back(a[i]);
    }
    sort(b);
    UNIQUE(b);
    for (int i = 0; i < len; ++i) {
        a[i] = lower_bound(all(b), a[i]) - b.begin();
    }
}

void compress(int a[], int len) {
    vi b;
    for (int i = 0; i < len; ++i) {
        b.push_back(a[i]);
    }
    sort(b);
    UNIQUE(b);
    for (int i = 0; i < len; ++i) {
        a[i] = lower_bound(all(b), a[i]) - b.begin();
    }
}
//要素が見つからなかったときに困る
#define binarySearch(a, v) (binary_search(all(a),v))
#define lowerIndex(a, v) (lower_bound(all(a),v)-a.begin())
#define lowerBound(a, v) (*lower_bound(all(a),v))
#define upperIndex(a, v) (upper_bound(all(a),v)-a.begin())
#define upperBound(a, v) (*upper_bound(all(a),v))
#define ans(a) cout<<a<<endl;continue;
#define poll(a) q.front();q.pop()
#define dpoll(a) q.front();q.pop_front()
#define pollLast(a) q.back();q.pop_back()
#define pollBack(a) q.back();q.pop_back()
template<class T> inline void fin(T s) { cout << s << endl, exit(0); }

template<class T> struct edge {
    int from, to;
    T cost;
    int id;
    int type;
    edge(int f, int t, T c = 1, int id = -1, int ty = -1) : from(f), to(t), cost(c), id(id), type(ty) {}
    bool operator<(const edge &b) const { return cost < b.cost; }
    bool operator>(const edge &b) const { return cost > b.cost; }
};

template<typename T> class graph {
protected:
    vector<bool> _used;
public :
    vector<vector<edge<T>>> g;
    vector<edge<T>> edges;
    int n;
    graph(int n) : n(n) { g.resize(n), _used.resize(n); }
    void clear() { g.clear(), edges.clear(); }
    void resize(int n) {
        this->n = n;
        g.resize(n);
        _used.resize(n);
    }
    int size() { return g.size(); }
    vector<edge<T> > &operator[](int i) { return g[i]; }
    virtual void add(int from, int to, T cost, int id, int ty) = 0;
    virtual bool used(edge<T> &e) = 0;
    virtual bool used(int id) = 0;
    virtual void del(edge<T> &e) = 0;
    virtual void del(int id) = 0;
};
template<class T> vector<T> dijkstra(const graph<T> &g, int s);
template<typename T =ll> class digraph : public graph<T> {
public:
    using graph<T>::g;
    using graph<T>::n;
    using graph<T>::edges;
    using graph<T>::_used;

    digraph(int n) : graph<T>(n) {}
    void add(int f, int t, T cost = 1, int id = -1, int ty = -1) {
        if (!(0 <= f && f < n && 0 <= t && t < n)) {
            debugline("digraph add");
            deb(f, t, cost, id, ty);
            ole();
        }
        if (id == -1)id = edges.size();
        g[f].emplace_back(f, t, cost, id, ty);
        edges.emplace_back(f, t, cost, id, ty);
    }
    bool used(edge<T> &e) { return _used[e.id]; }
    bool used(int id) { return _used[id]; }
    void del(edge<T> &e) { _used[e.id] = _used[e.id ^ 1] = 1; }
    void del(int id) { _used[id] = _used[id ^ 1] = 1; }
};
template<class T=int> class undigraph : public graph<T> {
public:
    using graph<T>::g;
    using graph<T>::n;
    using graph<T>::edges;
    using graph<T>::_used;
    undigraph(int n) : graph<T>(n) {
    }
    void add(int f, int t, T cost = 1, int id = -1, int ty = -1) {
        if (!(0 <= f && f < n && 0 <= t && t < n)) {
            debugline("undigraph add");
            deb(f, t, cost, id, ty);
            ole();
        }
        if (id == -1)id = edges.size();
        g[f].emplace_back(f, t, cost, id, ty);
        g[t].emplace_back(t, f, cost, id + 1, ty);
        edges.emplace_back(f, t, cost, id, ty);
        edges.emplace_back(t, f, cost, id + 1, ty);
    }
    void add(edge<T> &e) {
        int f = e.from, t = e.to, ty = e.type;
        T cost = e.cost;
        add(f, t, cost, ty);
    }
    bool used(edge<T> &e) { return _used[e.id]; }
    bool used(int id) { return _used[id]; }
    void del(edge<T> &e) { _used[e.id] = _used[e.id ^ 1] = 1; }
    void del(int id) { _used[id] = _used[id ^ 1] = 1; }
};


template<class T=int> class tree : public undigraph<T> {
public:
    using undigraph<T>::g;
    using undigraph<T>::n;
    using undigraph<T>::edges;
    using undigraph<T>::_used;
    vi dep;
    vi disv;
private:
    bool never = 1;
    int root = -1;
    vector<vector<int>> par;
    bool costallone;
    void dfs(int v, int p, int d) {
        dep[v] = d;
        par[0][v] = p;
        int lim = (*this)[v].size();
        for (int i = 0; i < lim; i++) {
            int t = g[v][i].to;
            if (t == p)con;
            dfs(t, v, d + 1);
        }
    }
    void built() {
        never = 0;
        n = g.size();
        par.assign(30, vi(n));
        dep.resize(n);
        costallone = 1;
        fora(e, edges)if (e.cost != 1)costallone = 0;
        dfs(root, -1, 0);
        rep(k, par.size() - 1) {
            rep(i, n) {
                if (par[k][i] == -1)par[k + 1][i] = -1;
                else par[k + 1][i] = par[k][par[k][i]];
            }
        }
        if (costallone)disv = dep;
        else disv = dijkstra(*this, root);
    }
    int _lca(int u, int v) {
        if (dep[u] > dep[v])swap(u, v);
        rep(k, par.size()) {
            if ((dep[u] - dep[v]) >> k & 1) {
                v = par[k][v];
            }
        }
        if (u == v)return u;
        rer(k, par.size() - 1) {
            if (par[k][u] != par[k][v]) {
                u = par[k][u];
                v = par[k][v];
            }
        }
        return par[0][u];
    }
    int _dis(int u, int v) {
        int p = _lca(u, v);
        return disv[u] + disv[v] - disv[p] * 2;
    }
public:
    tree(int n, int root = 0) : undigraph<T>(n), root(root) {}
    bool leaf(int v) {
        return sz(g[v]) == 1 && v != root;
    }
    int dis(int u, int v) {
        if (never) {
            built();
        }
        return _dis(u, v);
    }
    int lca(int u, int v) {
        if (never) {
            built();
        }
        return _lca(u, v);
    }
};
//辺によりメモリを大量消費
// よってedgesを消している
//頂点10^6でメモリを190MB(制限の8割)使う

//軽量化のため、辺を持たないbig gridクラスがあってもいいかもしれない
//
template<class T=int> class grid_k6 : public undigraph<T> {
public:
    using undigraph<T>::g;
    using undigraph<T>::n;
    using undigraph<T>::edges;
    using undigraph<T>::_used;
    int H, W;
    int eid = 0;
    void add(int f, int t, T cost = 1, int id = -1, int ty = -1) {
        if (!(0 <= f && f < n && 0 <= t && t < n)) {
            debugline("grid_k6 add");
            deb(f, t, cost, id, ty);
            ole();
        }
        g[f].emplace_back(f, t, cost, eid++, ty);
        g[t].emplace_back(t, f, cost, eid++, ty);
    }
    int getid(int h, int w) {
        if (!inside(h, w, H, W))return -1;
        return W * h + w;
    }
    P get2(int id) {
        return mp(id / W, id % W);
    }
    P operator()(int id) {
        return get2(id);
    }
    int operator()(int h, int w) {
        return getid(h, w);
    }
    grid_k6(int H, int W) : H(H), W(W), undigraph<T>(H * W) {
        rep(h, H) {
            rep(w, W) {
                int f = getid(h, w);
                if (w + 1 < W) add(f, getid(h, w + 1));
                if (h + 1 < H)add(f, getid(h + 1, w));
            }
        }
    }
    grid_k6(_vvc ba, char wall = '#') : H(sz(ba)), W(sz(ba[0])), undigraph<T>(sz(ba) * sz(ba[0])) {
        rep(h, H) {
            rep(w, W) {
                if (ba[h][w] == wall)con;
                int f = getid(h, w);
                if (w + 1 < W && ba[h][w + 1] != wall) {
                    add(f, getid(h, w + 1));
                }
                if (h + 1 < H && ba[h + 1][w] != wall) {
                    add(f, getid(h + 1, w));
                }
            }
        }
    }
    void add(int fh, int fw, int th, int tw) {
        add(getid(fh, fw), getid(th, tw));
    }
};

template<class T> bool nibu(const graph<T> &g) {
    if (g.edges.size() == 0)return true;
    UnionFind uf(g.n * 2);
    for (auto &&e :g.edges)uf.unite(e.from, e.to + g.n), uf.unite(e.from + g.n, e.to);
    rep(i, g.n)if (uf.same(i, i + g.n))return 0;
    return 1;
}
//二部グラフを色分けした際の頂点数を返す
template<class T> vp nibug(graph<T> &g) {
    vp cg;
    if (!nibu(g)) {
        debugline("nibu");
        ole();
    }
    int _n = g.size();
    vb _was(_n);
    queue<P> q;
    rep(i, _n) {
        if (_was[i])continue;
        q.push(mp(i, 1));
        _was[i] = 1;
        int red = 0;
        int coun = 0;
        while (q.size()) {
            int now = q.front().fi;
            int col = q.front().se;
            red += col;
            coun++;
            q.pop();
            forg(gi, g[now]) {
                if (_was[t])continue;
                q.push(mp(t, col ^ 1));
                _was[t] = 1;
            }
        }
        cg.push_back(mp(red, coun - red));
    }
    return cg;

}

template<class T> vector<T> dijkstra(const graph<T> &g, int s, int cant_arrive = -1) {
    if (!(0 <= s && s < g.n)) {
        debugline("dijkstra");
        deb(s, g.n);
        ole();
    }
    T initValue = MAX(T);
    vector<T> dis(g.n, initValue);
    priority_queue<pair<T, int>, vector<pair<T, int>>, greater<pair<T, int>>> q;
    dis[s] = 0;
    q.emplace(0, s);
    while (q.size()) {
        T nowc = q.top().fi;
        int i = q.top().se;
        q.pop();
        if (dis[i] != nowc)continue;
        for (auto &&e  : g.g[i]) {
            int to = e.to;
            T cost = nowc + e.cost;
            if (dis[to] > cost) {
                dis[to] = cost;
                q.emplace(dis[to], to);
            }
        }
    }
    //たどり着かないなら-1
    if (cant_arrive == -1)for (auto &&d :dis) if (d == initValue)d = -1;
    return dis;
}

//機能拡張
vp vtop(vi &a, vi &b) {
    vp res(sz(a));
    rep(i, sz(a))res[i] = mp(a[i], b[i]);
    return res;
}
void ptov(vp &p, vi &a, vi &b) {
    a.resize(sz(p));
    b.resize(sz(p));
    rep(i, sz(p))a[i] = p[i].fi, b[i] = p[i].se;
}
template<typename _CharT, typename _Traits, typename _Alloc>
basic_string<_CharT, _Traits, _Alloc>
operator+(const basic_string<_CharT, _Traits, _Alloc> &__lhs, const int __rv) {
    basic_string<_CharT, _Traits, _Alloc> __str(__lhs);
    __str.append(to_string(__rv));
    return __str;
}
template<typename _CharT, typename _Traits, typename _Alloc>
void operator+=(basic_string<_CharT, _Traits, _Alloc> &__lhs, const int __rv) {
    __lhs += to_string(__rv);
}
template<typename _CharT, typename _Traits, typename _Alloc>
basic_string<_CharT, _Traits, _Alloc>
operator+(const basic_string<_CharT, _Traits, _Alloc> &__lhs, const signed __rv) {
    basic_string<_CharT, _Traits, _Alloc> __str(__lhs);
    __str.append(to_string(__rv));
    return __str;
}
template<typename _CharT, typename _Traits, typename _Alloc>
void operator+=(basic_string<_CharT, _Traits, _Alloc> &__lhs, const signed __rv) {
    __lhs += to_string(__rv);
}
template<class T, class U> void operator+=(queue<T> &a, U v) {
    a.push(v);
}
template<class T, class U>
priority_queue<T, vector<T>, greater<T> > &operator+=(priority_queue<T, vector<T>, greater<T> > &a, U v) {
    a.push(v);
    return a;
}
template<class T, class U> priority_queue<T> &operator+=(priority_queue<T> &a, U v) {
    a.push(v);
    return a;
}
template<class T, class U> set<T> &operator+=(set<T> &a, U v) {
    a.insert(v);
    return a;
}
template<class T, class U> set<T, greater<T>> &operator+=(set<T, greater<T>> &a, U v) {
    a.insert(v);
    return a;
}

template<class T, class U> vector<T> &operator+=(vector<T> &a, U v) {
    a.pb(v);
    return a;
}
template<class T> vector<T> &operator+=(vector<T> &a, vector <T> &b) {
    a.pb(b);
    return a;
}
template<class T, class U> vector<T> &operator+=(vector<T> &a, initializer_list<U> v) {
    for (auto &&va :v)a.pb(va);
    return a;
}
template<class T> vector<T> &operator-=(vector<T> &a, vector <T> &b) {
    if (sz(a) != sz(b)) {
        debugline("vector<T> operator-=");
        deb(a);
        deb(b);
        ole();
    }
    rep(i, sz(a))a[i] -= b[i];
    return a;
}
template<class T> vector<T> &operator-(vector<T> &a, vector <T> &b) {
    if (sz(a) != sz(b)) {
        debugline("vector<T> operator-");
        deb(a);
        deb(b);
        ole();
    }
    vector<T> res;
    rep(i, sz(a))res[i] = a[i] - b[i];
    return res;
}


template<typename T> void remove(vector<T> &v, unsigned int i) { v.erase(v.begin() + i); }
template<typename T> void remove(vector<T> &v, unsigned int s, unsigned int e) {
    v.erase(v.begin() + s, v.begin() + e);
}
template<typename T> void removen(vector<T> &v, unsigned int s, unsigned int n) {
    v.erase(v.begin() + s, v.begin() + s + n);
}
template<typename T> void erase(vector<T> &v, unsigned int i) { v.erase(v.begin() + i); }
template<typename T> void erase(vector<T> &v, unsigned int s, unsigned int e) {
    v.erase(v.begin() + s, v.begin() + e);
}
template<typename T> void erasen(vector<T> &v, unsigned int s, unsigned int n) {
    v.erase(v.begin() + s, v.begin() + s + n);
}
template<typename T> void insert(vector<T> &v, unsigned int i, T t) { v.insert(v.begin() + i, t); }
template<typename T> void insert(vector<T> &v, unsigned int i, vector<T> list) {
    for (auto &&va :list)v.insert(v.begin() + i++, va);
}
template<typename T, typename U> void insert(vector<T> &v, initializer_list<U> list) {
    for (auto &&va :list)v.pb(va);
}
template<typename T, typename U> void insert(vector<T> &v, unsigned int i, initializer_list<U> list) {
    for (auto &&va :list)v.insert(v.begin() + i++, va);
}
template<typename T> void insert(set<T> &v, vector<T> list) {
    for (auto &&va :list)v.insert(va);
}
template<typename T> void insert(set<T> &v, initializer_list<T> list) {
    for (auto &&va :list)v.insert(va);
}


ll ma = numeric_limits<ll>::min();
ll mi = numeric_limits<ll>::max();
//閉路がなければtrue
bool topo(vi &res, digraph<int> &g) {
    int n = g.g.size();
    vi nyu(n);
    rep(i, n)for (auto &&e :g[i])nyu[e.to]++;
    queue<int> st;
    rep(i, n)if (nyu[i] == 0)st.push(i);
    while (st.size()) {
        int v = st.front();
        st.pop();
        res.pb(v);
        fora(e, g[v]) if (--nyu[e.to] == 0)st.push(e.to);
    }
    return res.size() == n;
}
//辞書順最小トポロジカルソート
bool topos(vi &res, digraph<int> &g) {
    int n = g.g.size();
    vi nyu(n);
    rep(i, n)for (auto &&e :g[i])nyu[e.to]++;
    //小さい順
    priority_queue<int, vector<int>, greater<int> > q;
    rep(i, n)if (nyu[i] == 0)q.push(i);
    while (q.size()) {
        int i = q.top();
        q.pop();
        res.pb(i);
        fora(e, g[i])if (--nyu[e.to] == 0)q.push(e.to);
    }
    return res.size() == n;
}
vector<string> split(const string a, const char deli) {
    string b = a + deli;
    int l = 0, r = 0, n = b.size();
    vector<string> res;
    rep(i, n) {
        if (b[i] == deli) {
            r = i;
            if (l < r)res.push_back(b.substr(l, r - l));
            l = i + 1;
        }
    }
    return res;
}
vector<string> split(const string a, const string deli) {
    string b = a + deli;
    int l = 0, r = 0, n = b.size(), dn = deli.size();
    vector<string> res;
    rep(i, n) {
        if (i + dn <= n && b.substr(i, i + dn) == deli) {
            r = i;
            if (l < r)res.push_back(b.substr(l, r - l));
            i += dn - 1;
            l = i + 1;
        }
    }
    return res;
}
void yn(bool a) {
    if (a)cout << "yes" << endl;
    else cout << "no" << endl;
}
void Yn(bool a) {
    if (a)cout << "Yes" << endl;
    else cout << "No" << endl;
}
void YN(bool a) {
    if (a)cout << "YES" << endl;
    else cout << "NO" << endl;
}
void fyn(bool a) {
    if (a)cout << "yes" << endl;
    else cout << "no" << endl;
    exit(0);
}
void fYn(bool a) {
    if (a)cout << "Yes" << endl;
    else cout << "No" << endl;
    exit(0);
}
void fYN(bool a) {
    if (a)cout << "YES" << endl;
    else cout << "NO" << endl;
    exit(0);
}
void Possible(bool a) {
    if (a)cout << "Possible" << endl;
    else cout << "Impossible" << endl;
    exit(0);
}

int n, m, k, d, H, W, x, y, z, q;
int cou;
vi a, b, c;
vvi (s, 0, 0);
vvc (ba, 0, 0);
vp p;
template<class T=string> struct rolling_hash_secure {
    const int b[4] = {11731, 11839, 12239, 14407};
    const int m[4] = {1000002571, 1000002631, 1000011277, 1000012187};
    T s;
    vector<long long> hash[4], pow[4];
    rolling_hash_secure(const T &S) : s(S) {
        int n = (int) S.size();
        for (int i = 0; i < 4; ++i) {
            hash[i].assign(n + 1, 0);
            pow[i].assign(n + 1, 1);
            for (int j = 0; j < n; ++j) {
                hash[i][j + 1] = (hash[i][j] * b[i] + S[j]) % m[i];
                pow[i][j + 1] = pow[i][j] * b[i] % m[i];
            }
        }
    }
    // get hash of S[left:right]
    inline ll get(int l, int r, int id) const {
        ll res = hash[id][r] - hash[id][l] * pow[id][r - l] % m[id];
        if (res < 0) res += m[id];
        return res;
    }
    inline P get(int l, int r) {
        return mp((get(l, r, 0) << 32) | get(l, r, 1), (get(l, r, 2) << 32) | get(l, r, 3));
    }
    inline P operator()(int l, int r) {
        return get(l, r);
    }
    // get lcp of S[a:] and S[b:]
    inline int lcp(int a, int b) const {
        int len = min((int) s.size() - a, (int) s.size() - b);
        int low = -1, high = len + 1;
        while (high - low > 1) {
            int mid = (low + high) / 2;
            if (get(a, a + mid, 0) != get(b, b + mid, 0)) high = mid;
            else if (get(a, a + mid, 1) != get(b, b + mid, 1)) high = mid;
            else low = mid;
        }
        return low;
    }
    // get lcp of S[a:] and T[b:]
    inline int lcp(const rolling_hash_secure &t, int a, int b) const {
        int len = min((int) s.size() - a, (int) s.size() - b);
        int low = -1, high = len + 1;
        while (high - low > 1) {
            int mid = (low + high) / 2;
            if (get(a, a + mid, 0) != get(b, b + mid, 0)) high = mid;
            else if (get(a, a + mid, 1) != get(b, b + mid, 1)) high = mid;
            else low = mid;
        }
        return low;
    }
};
void solve() {
    str s = sin();
    n = sz(s);
    rolling_hash_secure<> ro(s);
    setMod();
    umap<P, mint> dp;
    function<mint(int, int)> rec = [&](int l, int r) {
        P p = mp(l, r);
        if (dp.count(p))return dp[p];
        mint ret = 1;//[l r)
        rep(i, 1, n) {
            int ml = l + i;
            int mr = r - i;
            if (ml > mr)brk;
            if (ro(l, ml) == ro(mr, r))ret += rec(ml, mr);
        }
        return dp[p] = ret;
    };
    cout << rec(0, n) << endl;
}
int my(int n, vi a) {
    return 0;
}
int sister(int n, vi a) {
    return 0;
}

signed main() {
    solve();

#define _arg n,a
    //cin>>n;
    //na(a,n);
    //my(_arg);
    //cout << my(_arg) << endl;
#ifdef _DEBUG
    bool bad = 0;
    for (int i = 0, ok = 1; i < k5 && ok; i++) {
        int n = rand(1, 3);
        vi a = ranv(m, 1, 10);
        int myres = my(_arg);
        int res = sister(_arg);
        ok = myres == res;
        if (!ok) {
            cout << n << endl;
            cout << a << endl;
            cout << "正解 : " << res << endl;
            cout << "出力 : " << myres << endl;
            bad = 1;
            break;
        }
    }
//    if (!bad)cout << "お兄ちゃんすごい！天才！！" << endl;
#endif
    return 0;
};