#include <algorithm>
#include <array>
#include <bitset>
#include <cassert>
#include <chrono>
#include <cmath>
#include <complex>
#include <deque>
#include <forward_list>
#include <fstream>
#include <functional>
#include <iomanip>
#include <ios>
#include <iostream>
#include <limits>
#include <list>
#include <map>
#include <memory>
#include <numeric>
#include <optional>
#include <queue>
#include <random>
#include <set>
#include <sstream>
#include <stack>
#include <string>
#include <tuple>
#include <type_traits>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
using namespace std;
using lint = long long;
using pint = pair<int, int>;
using plint = pair<lint, lint>;
struct fast_ios { fast_ios(){ cin.tie(nullptr), ios::sync_with_stdio(false), cout << fixed << setprecision(20); }; } fast_ios_;
#define ALL(x) (x).begin(), (x).end()
#define FOR(i, begin, end) for(int i=(begin),i##_end_=(end);i<i##_end_;i++)
#define IFOR(i, begin, end) for(int i=(end)-1,i##_begin_=(begin);i>=i##_begin_;i--)
#define REP(i, n) FOR(i,0,n)
#define IREP(i, n) IFOR(i,0,n)
template <typename T> bool chmax(T &m, const T q) { return m < q ? (m = q, true) : false; }
template <typename T> bool chmin(T &m, const T q) { return m > q ? (m = q, true) : false; }
const std::vector<std::pair<int, int>> grid_dxs{{1, 0}, {-1, 0}, {0, 1}, {0, -1}};
int floor_lg(long long x) { return x <= 0 ? -1 : 63 - __builtin_clzll(x); }
template <class T1, class T2> T1 floor_div(T1 num, T2 den) { return (num > 0 ? num / den : -((-num + den - 1) / den)); }
template <class T1, class T2> std::pair<T1, T2> operator+(const std::pair<T1, T2> &l, const std::pair<T1, T2> &r) { return std::make_pair(l.first + r.first, l.second + r.second); }
template <class T1, class T2> std::pair<T1, T2> operator-(const std::pair<T1, T2> &l, const std::pair<T1, T2> &r) { return std::make_pair(l.first - r.first, l.second - r.second); }
template <class T> std::vector<T> sort_unique(std::vector<T> vec) { sort(vec.begin(), vec.end()), vec.erase(unique(vec.begin(), vec.end()), vec.end()); return vec; }
template <class T> int arglb(const std::vector<T> &v, const T &x) { return std::distance(v.begin(), std::lower_bound(v.begin(), v.end(), x)); }
template <class T> int argub(const std::vector<T> &v, const T &x) { return std::distance(v.begin(), std::upper_bound(v.begin(), v.end(), x)); }
template <class IStream, class T> IStream &operator>>(IStream &is, std::vector<T> &vec) { for (auto &v : vec) is >> v; return is; }

template <class OStream, class T> OStream &operator<<(OStream &os, const std::vector<T> &vec);
template <class OStream, class T, size_t sz> OStream &operator<<(OStream &os, const std::array<T, sz> &arr);
template <class OStream, class T, class TH> OStream &operator<<(OStream &os, const std::unordered_set<T, TH> &vec);
template <class OStream, class T, class U> OStream &operator<<(OStream &os, const pair<T, U> &pa);
template <class OStream, class T> OStream &operator<<(OStream &os, const std::deque<T> &vec);
template <class OStream, class T> OStream &operator<<(OStream &os, const std::set<T> &vec);
template <class OStream, class T> OStream &operator<<(OStream &os, const std::multiset<T> &vec);
template <class OStream, class T> OStream &operator<<(OStream &os, const std::unordered_multiset<T> &vec);
template <class OStream, class T, class U> OStream &operator<<(OStream &os, const std::pair<T, U> &pa);
template <class OStream, class TK, class TV> OStream &operator<<(OStream &os, const std::map<TK, TV> &mp);
template <class OStream, class TK, class TV, class TH> OStream &operator<<(OStream &os, const std::unordered_map<TK, TV, TH> &mp);
template <class OStream, class... T> OStream &operator<<(OStream &os, const std::tuple<T...> &tpl);

template <class OStream, class T> OStream &operator<<(OStream &os, const std::vector<T> &vec) { os << '['; for (auto v : vec) os << v << ','; os << ']'; return os; }
template <class OStream, class T, size_t sz> OStream &operator<<(OStream &os, const std::array<T, sz> &arr) { os << '['; for (auto v : arr) os << v << ','; os << ']'; return os; }
template <class... T> std::istream &operator>>(std::istream &is, std::tuple<T...> &tpl) { std::apply([&is](auto &&... args) { ((is >> args), ...);}, tpl); return is; }
template <class OStream, class... T> OStream &operator<<(OStream &os, const std::tuple<T...> &tpl) { os << '('; std::apply([&os](auto &&... args) { ((os << args << ','), ...);}, tpl); return os << ')'; }
template <class OStream, class T, class TH> OStream &operator<<(OStream &os, const std::unordered_set<T, TH> &vec) { os << '{'; for (auto v : vec) os << v << ','; os << '}'; return os; }
template <class OStream, class T> OStream &operator<<(OStream &os, const std::deque<T> &vec) { os << "deq["; for (auto v : vec) os << v << ','; os << ']'; return os; }
template <class OStream, class T> OStream &operator<<(OStream &os, const std::set<T> &vec) { os << '{'; for (auto v : vec) os << v << ','; os << '}'; return os; }
template <class OStream, class T> OStream &operator<<(OStream &os, const std::multiset<T> &vec) { os << '{'; for (auto v : vec) os << v << ','; os << '}'; return os; }
template <class OStream, class T> OStream &operator<<(OStream &os, const std::unordered_multiset<T> &vec) { os << '{'; for (auto v : vec) os << v << ','; os << '}'; return os; }
template <class OStream, class T, class U> OStream &operator<<(OStream &os, const std::pair<T, U> &pa) { return os << '(' << pa.first << ',' << pa.second << ')'; }
template <class OStream, class TK, class TV> OStream &operator<<(OStream &os, const std::map<TK, TV> &mp) { os << '{'; for (auto v : mp) os << v.first << "=>" << v.second << ','; os << '}'; return os; }
template <class OStream, class TK, class TV, class TH> OStream &operator<<(OStream &os, const std::unordered_map<TK, TV, TH> &mp) { os << '{'; for (auto v : mp) os << v.first << "=>" << v.second << ','; os << '}'; return os; }
#ifdef HITONANODE_LOCAL
const string COLOR_RESET = "\033[0m", BRIGHT_GREEN = "\033[1;32m", BRIGHT_RED = "\033[1;31m", BRIGHT_CYAN = "\033[1;36m", NORMAL_CROSSED = "\033[0;9;37m", RED_BACKGROUND = "\033[1;41m", NORMAL_FAINT = "\033[0;2m";
#define dbg(x) std::cerr << BRIGHT_CYAN << #x << COLOR_RESET << " = " << (x) << NORMAL_FAINT << " (L" << __LINE__ << ") " << __FILE__ << COLOR_RESET << std::endl
#define dbgif(cond, x) ((cond) ? std::cerr << BRIGHT_CYAN << #x << COLOR_RESET << " = " << (x) << NORMAL_FAINT << " (L" << __LINE__ << ") " << __FILE__ << COLOR_RESET << std::endl : std::cerr)
#else
#define dbg(x) ((void)0)
#define dbgif(cond, x) ((void)0)
#endif


#include <utility>
#include <vector>

// lowest common ancestor (LCA) for undirected weighted tree
template <typename T> struct UndirectedWeightedTree {
    int INVALID = -1;
    int V, lgV;
    int E;
    int root;
    std::vector<std::vector<std::pair<int, int>>> adj; // (nxt_vertex, edge_id)
    // vector<pint> edge; // edges[edge_id] = (vertex_id, vertex_id)
    std::vector<T> weight;     // w[edge_id]
    std::vector<int> par;      // parent_vertex_id[vertex_id]
    std::vector<int> depth;    // depth_from_root[vertex_id]
    std::vector<T> acc_weight; // w_sum_from_root[vertex_id]

    void _fix_root_dfs(int now, int prv, int prv_edge_id) {
        par[now] = prv;
        if (prv_edge_id != INVALID) acc_weight[now] = acc_weight[prv] + weight[prv_edge_id];
        for (auto nxt : adj[now])
            if (nxt.first != prv) {
                depth[nxt.first] = depth[now] + 1;
                _fix_root_dfs(nxt.first, now, nxt.second);
            }
    }

    UndirectedWeightedTree() = default;
    UndirectedWeightedTree(int N) : V(N), E(0), adj(N) {
        lgV = 1;
        while (1 << lgV < V) lgV++;
    }

    void add_edge(int u, int v, T w) {
        adj[u].emplace_back(v, E);
        adj[v].emplace_back(u, E);
        // edge.emplace_back(u, v);
        weight.emplace_back(w);
        E++;
    }

    std::vector<std::vector<int>> doubling;
    void _doubling_precalc() {
        doubling.assign(lgV, std::vector<int>(V));
        doubling[0] = par;
        for (int d = 0; d < lgV - 1; d++)
            for (int i = 0; i < V; i++) {
                if (doubling[d][i] == INVALID)
                    doubling[d + 1][i] = INVALID;
                else
                    doubling[d + 1][i] = doubling[d][doubling[d][i]];
            }
    }

    void fix_root(int r) {
        root = r;
        par.resize(V);
        depth.resize(V);
        depth[r] = 0;
        acc_weight.resize(V);
        acc_weight[r] = 0;
        _fix_root_dfs(root, INVALID, INVALID);
        _doubling_precalc();
    }

    int kth_parent(int x, int k) const {
        if (depth[x] < k) return INVALID;
        for (int d = 0; d < lgV; d++) {
            if (x == INVALID) return INVALID;
            if (k & (1 << d)) x = doubling[d][x];
        }
        return x;
    }

    int lowest_common_ancestor(int u, int v) const {
        if (depth[u] > depth[v]) std::swap(u, v);

        v = kth_parent(v, depth[v] - depth[u]);
        if (u == v) return u;
        for (int d = lgV - 1; d >= 0; d--) {
            if (doubling[d][u] != doubling[d][v]) u = doubling[d][u], v = doubling[d][v];
        }
        return par[u];
    }

    T path_length(int u, int v) const {
        // Not distance, but the sum of weights
        int r = lowest_common_ancestor(u, v);
        return (acc_weight[u] - acc_weight[r]) + (acc_weight[v] - acc_weight[r]);
    }

    int s_to_t_by_k_steps(int s, int t, int k) const {
        int l = lowest_common_ancestor(s, t);
        int dsl = depth[s] - depth[l], dtl = depth[t] - depth[l];
        if (k > dsl + dtl) {
            return INVALID;
        } else if (k < dsl) {
            return kth_parent(s, k);
        } else if (k == dsl) {
            return l;
        } else {
            return kth_parent(t, dsl + dtl - k);
        }
    }
};

#include <algorithm>
#include <cassert>
#include <deque>
#include <fstream>
#include <functional>
#include <limits>
#include <queue>
#include <string>
#include <tuple>
#include <utility>
#include <vector>

template <typename T, T INF = std::numeric_limits<T>::max() / 2, int INVALID = -1>
struct shortest_path {
    int V, E;
    bool single_positive_weight;
    T wmin, wmax;

    std::vector<std::pair<int, T>> tos;
    std::vector<int> head;
    std::vector<std::tuple<int, int, T>> edges;

    void build_() {
        if (int(tos.size()) == E and int(head.size()) == V + 1) return;
        tos.resize(E);
        head.assign(V + 1, 0);
        for (const auto &e : edges) ++head[std::get<0>(e) + 1];
        for (int i = 0; i < V; ++i) head[i + 1] += head[i];
        auto cur = head;
        for (const auto &e : edges) {
            tos[cur[std::get<0>(e)]++] = std::make_pair(std::get<1>(e), std::get<2>(e));
        }
    }

    shortest_path(int V = 0) : V(V), E(0), single_positive_weight(true), wmin(0), wmax(0) {}
    void add_edge(int s, int t, T w) {
        assert(0 <= s and s < V);
        assert(0 <= t and t < V);
        edges.emplace_back(s, t, w);
        ++E;
        if (w > 0 and wmax > 0 and wmax != w) single_positive_weight = false;
        wmin = std::min(wmin, w);
        wmax = std::max(wmax, w);
    }

    void add_bi_edge(int u, int v, T w) {
        add_edge(u, v, w);
        add_edge(v, u, w);
    }

    std::vector<T> dist;
    std::vector<int> prev;

    // Dijkstra algorithm
    // - Requirement: wmin >= 0
    // - Complexity: O(E log E)
    using Pque = std::priority_queue<std::pair<T, int>, std::vector<std::pair<T, int>>,
                                     std::greater<std::pair<T, int>>>;
    template <class Heap = Pque> void dijkstra(int s, int t = INVALID) {
        assert(0 <= s and s < V);
        build_();
        dist.assign(V, INF);
        prev.assign(V, INVALID);
        dist[s] = 0;
        Heap pq;
        pq.emplace(0, s);
        while (!pq.empty()) {
            T d;
            int v;
            std::tie(d, v) = pq.top();
            pq.pop();
            if (t == v) return;
            if (dist[v] < d) continue;
            for (int e = head[v]; e < head[v + 1]; ++e) {
                const auto &nx = tos[e];
                T dnx = d + nx.second;
                if (dist[nx.first] > dnx) {
                    dist[nx.first] = dnx, prev[nx.first] = v;
                    pq.emplace(dnx, nx.first);
                }
            }
        }
    }

    // Dijkstra algorithm
    // - Requirement: wmin >= 0
    // - Complexity: O(V^2 + E)
    void dijkstra_vquad(int s, int t = INVALID) {
        assert(0 <= s and s < V);
        build_();
        dist.assign(V, INF);
        prev.assign(V, INVALID);
        dist[s] = 0;
        std::vector<char> fixed(V, false);
        while (true) {
            int r = INVALID;
            T dr = INF;
            for (int i = 0; i < V; i++) {
                if (!fixed[i] and dist[i] < dr) r = i, dr = dist[i];
            }
            if (r == INVALID or r == t) break;
            fixed[r] = true;
            int nxt;
            T dx;
            for (int e = head[r]; e < head[r + 1]; ++e) {
                std::tie(nxt, dx) = tos[e];
                if (dist[nxt] > dist[r] + dx) dist[nxt] = dist[r] + dx, prev[nxt] = r;
            }
        }
    }

    // Bellman-Ford algorithm
    // - Requirement: no negative loop
    // - Complexity: O(VE)
    bool bellman_ford(int s, int nb_loop) {
        assert(0 <= s and s < V);
        build_();
        dist.assign(V, INF), prev.assign(V, INVALID);
        dist[s] = 0;
        for (int l = 0; l < nb_loop; l++) {
            bool upd = false;
            for (int v = 0; v < V; v++) {
                if (dist[v] == INF) continue;
                for (int e = head[v]; e < head[v + 1]; ++e) {
                    const auto &nx = tos[e];
                    T dnx = dist[v] + nx.second;
                    if (dist[nx.first] > dnx) dist[nx.first] = dnx, prev[nx.first] = v, upd = true;
                }
            }
            if (!upd) return true;
        }
        return false;
    }

    // Bellman-ford algorithm using deque
    // - Requirement: no negative loop
    // - Complexity: O(VE)
    void spfa(int s) {
        assert(0 <= s and s < V);
        build_();
        dist.assign(V, INF);
        prev.assign(V, INVALID);
        dist[s] = 0;
        std::deque<int> q;
        std::vector<char> in_queue(V);
        q.push_back(s), in_queue[s] = 1;
        while (!q.empty()) {
            int now = q.front();
            q.pop_front(), in_queue[now] = 0;
            for (int e = head[now]; e < head[now + 1]; ++e) {
                const auto &nx = tos[e];
                T dnx = dist[now] + nx.second;
                int nxt = nx.first;
                if (dist[nxt] > dnx) {
                    dist[nxt] = dnx;
                    if (!in_queue[nxt]) {
                        if (q.size() and dnx < dist[q.front()]) { // Small label first optimization
                            q.push_front(nxt);
                        } else {
                            q.push_back(nxt);
                        }
                        prev[nxt] = now, in_queue[nxt] = 1;
                    }
                }
            }
        }
    }

    // 01-BFS
    // - Requirement: all weights must be 0 or w (positive constant).
    // - Complexity: O(V + E)
    void zero_one_bfs(int s, int t = INVALID) {
        assert(0 <= s and s < V);
        build_();
        dist.assign(V, INF), prev.assign(V, INVALID);
        dist[s] = 0;
        std::vector<int> q(V * 4);
        int ql = V * 2, qr = V * 2;
        q[qr++] = s;
        while (ql < qr) {
            int v = q[ql++];
            if (v == t) return;
            for (int e = head[v]; e < head[v + 1]; ++e) {
                const auto &nx = tos[e];
                T dnx = dist[v] + nx.second;
                if (dist[nx.first] > dnx) {
                    dist[nx.first] = dnx, prev[nx.first] = v;
                    if (nx.second) {
                        q[qr++] = nx.first;
                    } else {
                        q[--ql] = nx.first;
                    }
                }
            }
        }
    }

    // Dial's algorithm
    // - Requirement: wmin >= 0
    // - Complexity: O(wmax * V + E)
    void dial(int s, int t = INVALID) {
        assert(0 <= s and s < V);
        build_();
        dist.assign(V, INF), prev.assign(V, INVALID);
        dist[s] = 0;
        std::vector<std::vector<std::pair<int, T>>> q(wmax + 1);
        q[0].emplace_back(s, dist[s]);
        int ninq = 1;

        int cur = 0;
        T dcur = 0;
        for (; ninq; ++cur, ++dcur) {
            if (cur == wmax + 1) cur = 0;
            while (!q[cur].empty()) {
                int v = q[cur].back().first;
                T dnow = q[cur].back().second;
                q[cur].pop_back(), --ninq;
                if (v == t) return;
                if (dist[v] < dnow) continue;

                for (int e = head[v]; e < head[v + 1]; ++e) {
                    const auto &nx = tos[e];
                    T dnx = dist[v] + nx.second;
                    if (dist[nx.first] > dnx) {
                        dist[nx.first] = dnx, prev[nx.first] = v;
                        int nxtcur = cur + int(nx.second);
                        if (nxtcur >= int(q.size())) nxtcur -= q.size();
                        q[nxtcur].emplace_back(nx.first, dnx), ++ninq;
                    }
                }
            }
        }
    }

    // Solver for DAG
    // - Requirement: graph is DAG
    // - Complexity: O(V + E)
    bool dag_solver(int s) {
        assert(0 <= s and s < V);
        build_();
        dist.assign(V, INF), prev.assign(V, INVALID);
        dist[s] = 0;
        std::vector<int> indeg(V, 0);
        std::vector<int> q(V * 2);
        int ql = 0, qr = 0;
        q[qr++] = s;
        while (ql < qr) {
            int now = q[ql++];
            for (int e = head[now]; e < head[now + 1]; ++e) {
                const auto &nx = tos[e];
                ++indeg[nx.first];
                if (indeg[nx.first] == 1) q[qr++] = nx.first;
            }
        }
        ql = qr = 0;
        q[qr++] = s;
        while (ql < qr) {
            int now = q[ql++];
            for (int e = head[now]; e < head[now + 1]; ++e) {
                const auto &nx = tos[e];
                --indeg[nx.first];
                if (dist[nx.first] > dist[now] + nx.second)
                    dist[nx.first] = dist[now] + nx.second, prev[nx.first] = now;
                if (indeg[nx.first] == 0) q[qr++] = nx.first;
            }
        }
        return *max_element(indeg.begin(), indeg.end()) == 0;
    }

    // Retrieve a sequence of vertex ids that represents shortest path [s, ..., goal]
    // If not reachable to goal, return {}
    std::vector<int> retrieve_path(int goal) const {
        assert(int(prev.size()) == V);
        assert(0 <= goal and goal < V);
        if (dist[goal] == INF) return {};
        std::vector<int> ret{goal};
        while (prev[goal] != INVALID) {
            goal = prev[goal];
            ret.push_back(goal);
        }
        std::reverse(ret.begin(), ret.end());
        return ret;
    }

    void solve(int s, int t = INVALID) {
        if (wmin >= 0) {
            if (single_positive_weight) {
                zero_one_bfs(s, t);
            } else if (wmax <= 10) {
                dial(s, t);
            } else {
                if ((long long)V * V < (E << 4)) {
                    dijkstra_vquad(s, t);
                } else {
                    dijkstra(s, t);
                }
            }
        } else {
            bellman_ford(s, V);
        }
    }

    // Warshall-Floyd algorithm
    // - Requirement: no negative loop
    // - Complexity: O(E + V^3)
    std::vector<std::vector<T>> floyd_warshall() {
        build_();
        std::vector<std::vector<T>> dist2d(V, std::vector<T>(V, INF));
        for (int i = 0; i < V; i++) {
            dist2d[i][i] = 0;
            for (const auto &e : edges) {
                int s = std::get<0>(e), t = std::get<1>(e);
                dist2d[s][t] = std::min(dist2d[s][t], std::get<2>(e));
            }
        }
        for (int k = 0; k < V; k++) {
            for (int i = 0; i < V; i++) {
                if (dist2d[i][k] == INF) continue;
                for (int j = 0; j < V; j++) {
                    if (dist2d[k][j] == INF) continue;
                    dist2d[i][j] = std::min(dist2d[i][j], dist2d[i][k] + dist2d[k][j]);
                }
            }
        }
        return dist2d;
    }

    void to_dot(std::string filename = "shortest_path") const {
        std::ofstream ss(filename + ".DOT");
        ss << "digraph{\n";
        build_();
        for (int i = 0; i < V; i++) {
            for (int e = head[i]; e < head[i + 1]; ++e) {
                ss << i << "->" << tos[e].first << "[label=" << tos[e].second << "];\n";
            }
        }
        ss << "}\n";
        ss.close();
        return;
    }
};

#include <cassert>
#include <iostream>
#include <set>
#include <vector>

template <int md> struct ModInt {
    static_assert(md > 1);
    using lint = long long;
    constexpr static int mod() { return md; }
    static int get_primitive_root() {
        static int primitive_root = 0;
        if (!primitive_root) {
            primitive_root = [&]() {
                std::set<int> fac;
                int v = md - 1;
                for (lint i = 2; i * i <= v; i++)
                    while (v % i == 0) fac.insert(i), v /= i;
                if (v > 1) fac.insert(v);
                for (int g = 1; g < md; g++) {
                    bool ok = true;
                    for (auto i : fac)
                        if (ModInt(g).pow((md - 1) / i) == 1) {
                            ok = false;
                            break;
                        }
                    if (ok) return g;
                }
                return -1;
            }();
        }
        return primitive_root;
    }
    int val_;
    int val() const noexcept { return val_; }
    constexpr ModInt() : val_(0) {}
    constexpr ModInt &_setval(lint v) { return val_ = (v >= md ? v - md : v), *this; }
    constexpr ModInt(lint v) { _setval(v % md + md); }
    constexpr explicit operator bool() const { return val_ != 0; }
    constexpr ModInt operator+(const ModInt &x) const {
        return ModInt()._setval((lint)val_ + x.val_);
    }
    constexpr ModInt operator-(const ModInt &x) const {
        return ModInt()._setval((lint)val_ - x.val_ + md);
    }
    constexpr ModInt operator*(const ModInt &x) const {
        return ModInt()._setval((lint)val_ * x.val_ % md);
    }
    constexpr ModInt operator/(const ModInt &x) const {
        return ModInt()._setval((lint)val_ * x.inv().val() % md);
    }
    constexpr ModInt operator-() const { return ModInt()._setval(md - val_); }
    constexpr ModInt &operator+=(const ModInt &x) { return *this = *this + x; }
    constexpr ModInt &operator-=(const ModInt &x) { return *this = *this - x; }
    constexpr ModInt &operator*=(const ModInt &x) { return *this = *this * x; }
    constexpr ModInt &operator/=(const ModInt &x) { return *this = *this / x; }
    friend constexpr ModInt operator+(lint a, const ModInt &x) { return ModInt(a) + x; }
    friend constexpr ModInt operator-(lint a, const ModInt &x) { return ModInt(a) - x; }
    friend constexpr ModInt operator*(lint a, const ModInt &x) { return ModInt(a) * x; }
    friend constexpr ModInt operator/(lint a, const ModInt &x) { return ModInt(a) / x; }
    constexpr bool operator==(const ModInt &x) const { return val_ == x.val_; }
    constexpr bool operator!=(const ModInt &x) const { return val_ != x.val_; }
    constexpr bool operator<(const ModInt &x) const {
        return val_ < x.val_;
    } // To use std::map<ModInt, T>
    friend std::istream &operator>>(std::istream &is, ModInt &x) {
        lint t;
        return is >> t, x = ModInt(t), is;
    }
    constexpr friend std::ostream &operator<<(std::ostream &os, const ModInt &x) {
        return os << x.val_;
    }

    constexpr ModInt pow(lint n) const {
        ModInt ans = 1, tmp = *this;
        while (n) {
            if (n & 1) ans *= tmp;
            tmp *= tmp, n >>= 1;
        }
        return ans;
    }

    static constexpr int cache_limit = std::min(md, 1 << 21);
    static std::vector<ModInt> facs, facinvs, invs;

    constexpr static void _precalculation(int N) {
        const int l0 = facs.size();
        if (N > md) N = md;
        if (N <= l0) return;
        facs.resize(N), facinvs.resize(N), invs.resize(N);
        for (int i = l0; i < N; i++) facs[i] = facs[i - 1] * i;
        facinvs[N - 1] = facs.back().pow(md - 2);
        for (int i = N - 2; i >= l0; i--) facinvs[i] = facinvs[i + 1] * (i + 1);
        for (int i = N - 1; i >= l0; i--) invs[i] = facinvs[i] * facs[i - 1];
    }

    constexpr ModInt inv() const {
        if (this->val_ < cache_limit) {
            if (facs.empty()) facs = {1}, facinvs = {1}, invs = {0};
            while (this->val_ >= int(facs.size())) _precalculation(facs.size() * 2);
            return invs[this->val_];
        } else {
            return this->pow(md - 2);
        }
    }

    constexpr static ModInt fac(int n) {
        assert(n >= 0);
        if (n >= md) return ModInt(0);
        while (n >= int(facs.size())) _precalculation(facs.size() * 2);
        return facs[n];
    }

    constexpr static ModInt facinv(int n) {
        assert(n >= 0);
        if (n >= md) return ModInt(0);
        while (n >= int(facs.size())) _precalculation(facs.size() * 2);
        return facinvs[n];
    }

    constexpr static ModInt doublefac(int n) {
        assert(n >= 0);
        if (n >= md) return ModInt(0);
        long long k = (n + 1) / 2;
        return (n & 1) ? ModInt::fac(k * 2) / (ModInt(2).pow(k) * ModInt::fac(k))
                       : ModInt::fac(k) * ModInt(2).pow(k);
    }

    constexpr static ModInt nCr(int n, int r) {
        assert(n >= 0);
        if (r < 0 or n < r) return ModInt(0);
        return ModInt::fac(n) * ModInt::facinv(r) * ModInt::facinv(n - r);
    }

    constexpr static ModInt nPr(int n, int r) {
        assert(n >= 0);
        if (r < 0 or n < r) return ModInt(0);
        return ModInt::fac(n) * ModInt::facinv(n - r);
    }

    static ModInt binom(long long n, long long r) {
        static long long bruteforce_times = 0;

        if (r < 0 or n < r) return ModInt(0);
        if (n <= bruteforce_times or n < (int)facs.size()) return ModInt::nCr(n, r);

        r = std::min(r, n - r);
        assert((int)r == r);

        ModInt ret = ModInt::facinv(r);
        for (int i = 0; i < r; ++i) ret *= n - i;
        bruteforce_times += r;

        return ret;
    }

    // Multinomial coefficient, (k_1 + k_2 + ... + k_m)! / (k_1! k_2! ... k_m!)
    // Complexity: O(sum(ks))
    // Verify: https://yukicoder.me/problems/no/3178
    template <class Vec> static ModInt multinomial(const Vec &ks) {
        ModInt ret{1};
        int sum = 0;
        for (int k : ks) {
            assert(k >= 0);
            ret *= ModInt::facinv(k), sum += k;
        }
        return ret * ModInt::fac(sum);
    }
    template <class... Args> static ModInt multinomial(Args... args) {
        int sum = (0 + ... + args);
        ModInt result = (1 * ... * ModInt::facinv(args));
        return ModInt::fac(sum) * result;
    }

    // Catalan number, C_n = binom(2n, n) / (n + 1) = # of Dyck words of length 2n
    // C_0 = 1, C_1 = 1, C_2 = 2, C_3 = 5, C_4 = 14, ...
    // https://oeis.org/A000108
    // Complexity: O(n)
    static ModInt catalan(int n) {
        if (n < 0) return ModInt(0);
        return ModInt::fac(n * 2) * ModInt::facinv(n + 1) * ModInt::facinv(n);
    }

    ModInt sqrt() const {
        if (val_ == 0) return 0;
        if (md == 2) return val_;
        if (pow((md - 1) / 2) != 1) return 0;
        ModInt b = 1;
        while (b.pow((md - 1) / 2) == 1) b += 1;
        int e = 0, m = md - 1;
        while (m % 2 == 0) m >>= 1, e++;
        ModInt x = pow((m - 1) / 2), y = (*this) * x * x;
        x *= (*this);
        ModInt z = b.pow(m);
        while (y != 1) {
            int j = 0;
            ModInt t = y;
            while (t != 1) j++, t *= t;
            z = z.pow(1LL << (e - j - 1));
            x *= z, z *= z, y *= z;
            e = j;
        }
        return ModInt(std::min(x.val_, md - x.val_));
    }
};
template <int md> std::vector<ModInt<md>> ModInt<md>::facs = {1};
template <int md> std::vector<ModInt<md>> ModInt<md>::facinvs = {1};
template <int md> std::vector<ModInt<md>> ModInt<md>::invs = {0};

using mint = ModInt<998244353>;

#include <algorithm>
#include <array>
#include <cassert>
#include <tuple>
#include <vector>

// Integer convolution for arbitrary mod
// with NTT (and Garner's algorithm) for ModInt / ModIntRuntime class.
// We skip Garner's algorithm if `skip_garner` is true or mod is in `nttprimes`.
// input: a (size: n), b (size: m)
// return: vector (size: n + m - 1)
template <typename MODINT>
std::vector<MODINT> nttconv(std::vector<MODINT> a, std::vector<MODINT> b, bool skip_garner);

constexpr int nttprimes[3] = {998244353, 167772161, 469762049};

// Integer FFT (Fast Fourier Transform) for ModInt class
// (Also known as Number Theoretic Transform, NTT)
// is_inverse: inverse transform
// ** Input size must be 2^n **
template <typename MODINT> void ntt(std::vector<MODINT> &a, bool is_inverse = false) {
    int n = a.size();
    if (n == 1) return;
    static const int mod = MODINT::mod();
    static const MODINT root = MODINT::get_primitive_root();
    assert(__builtin_popcount(n) == 1 and (mod - 1) % n == 0);

    static std::vector<MODINT> w{1}, iw{1};
    for (int m = w.size(); m < n / 2; m *= 2) {
        MODINT dw = root.pow((mod - 1) / (4 * m)), dwinv = 1 / dw;
        w.resize(m * 2), iw.resize(m * 2);
        for (int i = 0; i < m; i++) w[m + i] = w[i] * dw, iw[m + i] = iw[i] * dwinv;
    }

    if (!is_inverse) {
        for (int m = n; m >>= 1;) {
            for (int s = 0, k = 0; s < n; s += 2 * m, k++) {
                for (int i = s; i < s + m; i++) {
                    MODINT x = a[i], y = a[i + m] * w[k];
                    a[i] = x + y, a[i + m] = x - y;
                }
            }
        }
    } else {
        for (int m = 1; m < n; m *= 2) {
            for (int s = 0, k = 0; s < n; s += 2 * m, k++) {
                for (int i = s; i < s + m; i++) {
                    MODINT x = a[i], y = a[i + m];
                    a[i] = x + y, a[i + m] = (x - y) * iw[k];
                }
            }
        }
        int n_inv = MODINT(n).inv().val();
        for (auto &v : a) v *= n_inv;
    }
}
template <int MOD>
std::vector<ModInt<MOD>> nttconv_(const std::vector<int> &a, const std::vector<int> &b) {
    int sz = a.size();
    assert(a.size() == b.size() and __builtin_popcount(sz) == 1);
    std::vector<ModInt<MOD>> ap(sz), bp(sz);
    for (int i = 0; i < sz; i++) ap[i] = a[i], bp[i] = b[i];
    ntt(ap, false);
    if (a == b)
        bp = ap;
    else
        ntt(bp, false);
    for (int i = 0; i < sz; i++) ap[i] *= bp[i];
    ntt(ap, true);
    return ap;
}
long long garner_ntt_(int r0, int r1, int r2, int mod) {
    using mint2 = ModInt<nttprimes[2]>;
    static const long long m01 = 1LL * nttprimes[0] * nttprimes[1];
    static const long long m0_inv_m1 = ModInt<nttprimes[1]>(nttprimes[0]).inv().val();
    static const long long m01_inv_m2 = mint2(m01).inv().val();

    int v1 = (m0_inv_m1 * (r1 + nttprimes[1] - r0)) % nttprimes[1];
    auto v2 = (mint2(r2) - r0 - mint2(nttprimes[0]) * v1) * m01_inv_m2;
    return (r0 + 1LL * nttprimes[0] * v1 + m01 % mod * v2.val()) % mod;
}
template <typename MODINT>
std::vector<MODINT> nttconv(std::vector<MODINT> a, std::vector<MODINT> b, bool skip_garner) {
    if (a.empty() or b.empty()) return {};
    int sz = 1, n = a.size(), m = b.size();
    while (sz < n + m) sz <<= 1;
    if (sz <= 16) {
        std::vector<MODINT> ret(n + m - 1);
        for (int i = 0; i < n; i++) {
            for (int j = 0; j < m; j++) ret[i + j] += a[i] * b[j];
        }
        return ret;
    }
    int mod = MODINT::mod();
    if (skip_garner or
        std::find(std::begin(nttprimes), std::end(nttprimes), mod) != std::end(nttprimes)) {
        a.resize(sz), b.resize(sz);
        if (a == b) {
            ntt(a, false);
            b = a;
        } else {
            ntt(a, false), ntt(b, false);
        }
        for (int i = 0; i < sz; i++) a[i] *= b[i];
        ntt(a, true);
        a.resize(n + m - 1);
    } else {
        std::vector<int> ai(sz), bi(sz);
        for (int i = 0; i < n; i++) ai[i] = a[i].val();
        for (int i = 0; i < m; i++) bi[i] = b[i].val();
        auto ntt0 = nttconv_<nttprimes[0]>(ai, bi);
        auto ntt1 = nttconv_<nttprimes[1]>(ai, bi);
        auto ntt2 = nttconv_<nttprimes[2]>(ai, bi);
        a.resize(n + m - 1);
        for (int i = 0; i < n + m - 1; i++)
            a[i] = garner_ntt_(ntt0[i].val(), ntt1[i].val(), ntt2[i].val(), mod);
    }
    return a;
}

template <typename MODINT>
std::vector<MODINT> nttconv(const std::vector<MODINT> &a, const std::vector<MODINT> &b) {
    return nttconv<MODINT>(a, b, false);
}


int main() {
    int N, S, T;
    cin >> N >> S >> T;
    --S, --T;
    UndirectedWeightedTree<int> tree(N);
    shortest_path<int> sp(N);
    vector<pint> edges(N - 1);

    vector<int> deg(N);
    for (auto &[a, b] : edges) {
        cin >> a >> b;
        --a, --b;
        tree.add_edge(a, b, 1);
        sp.add_bi_edge(a, b, 1);
        deg.at(a)++;
        deg.at(b)++;
    }
    dbg(edges);

    tree.fix_root(0);

    auto [s0, s1] = edges.at(S);
    auto [t0, t1] = edges.at(T);

    if (tree.path_length(s0, t0) > tree.path_length(s1, t0)) swap(s0, s1);
    if (tree.path_length(s0, t0) > tree.path_length(s0, t1)) swap(t0, t1);

    sp.solve(s0, t0);
    const auto path = sp.retrieve_path(t0);
    dbg(path);

    vector<int> dseq;
    for (int p : path) dseq.push_back(deg.at(p) - 2);
    dbg(dseq);

    auto rec = [&](auto &&self, int l, int r) -> vector<mint> {
        if (l + 1 == r) {
            const int n = dseq.at(l);
            vector<mint> ret(n + 2);
            REP(k, n + 1) ret.at(k + 1) = mint::nPr(n, k);
            return ret;
        }
        const int mid = (l + r) / 2;
        return nttconv(self(self, l, mid), self(self, mid, r));
    };

    auto ret = rec(rec, 0, dseq.size());
    ret.resize(N);
    dbg(ret);
    for (auto e : ret) cout << e << ' ';
    cout << '\n';
}