結果
| 問題 | No.2587 Random Walk on Tree |
| ユーザー |
 NyaanNyaan
|
| 提出日時 | 2023-12-15 02:07:51 |
| 言語 | C++17 (gcc 12.3.0 + boost 1.83.0) |
| 結果 |
RE
|
| 実行時間 | - |
| コード長 | 56,824 bytes |
| コンパイル時間 | 5,894 ms |
| コンパイル使用メモリ | 331,772 KB |
| 実行使用メモリ | 45,392 KB |
| 最終ジャッジ日時 | 2024-09-27 12:56:00 |
| 合計ジャッジ時間 | 28,337 ms |
|
ジャッジサーバーID （参考情報） |
judge5 / judge4 |
（要ログイン）
テストケース
テストケース表示| 入力 | 結果 | 実行時間 実行使用メモリ |
|---|---|---|
| testcase_00 | RE | - |
| testcase_01 | AC | 2 ms
6,940 KB |
| testcase_02 | RE | - |
| testcase_03 | RE | - |
| testcase_04 | RE | - |
| testcase_05 | RE | - |
| testcase_06 | RE | - |
| testcase_07 | RE | - |
| testcase_08 | RE | - |
| testcase_09 | RE | - |
| testcase_10 | RE | - |
| testcase_11 | RE | - |
| testcase_12 | RE | - |
| testcase_13 | RE | - |
| testcase_14 | RE | - |
| testcase_15 | RE | - |
| testcase_16 | RE | - |
| testcase_17 | RE | - |
| testcase_18 | RE | - |
| testcase_19 | TLE | - |
| testcase_20 | -- | - |
| testcase_21 | -- | - |
| testcase_22 | -- | - |
| testcase_23 | -- | - |
| testcase_24 | -- | - |
| testcase_25 | -- | - |
| testcase_26 | -- | - |
| testcase_27 | -- | - |
| testcase_28 | -- | - |
| testcase_29 | -- | - |
| testcase_30 | -- | - |
| testcase_31 | -- | - |
| testcase_32 | -- | - |
| testcase_33 | -- | - |
| testcase_34 | -- | - |
| testcase_35 | -- | - |
| testcase_36 | -- | - |
| testcase_37 | -- | - |
| testcase_38 | -- | - |
| testcase_39 | -- | - |
ソースコード
/**
* date : 2023-12-15 02:07:41
* author : Nyaan
*/
#define NDEBUG
using namespace std;
// intrinstic
#include <immintrin.h>
#include <algorithm>
#include <array>
#include <bitset>
#include <cassert>
#include <cctype>
#include <cfenv>
#include <cfloat>
#include <chrono>
#include <cinttypes>
#include <climits>
#include <cmath>
#include <complex>
#include <cstdarg>
#include <cstddef>
#include <cstdint>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <deque>
#include <fstream>
#include <functional>
#include <initializer_list>
#include <iomanip>
#include <ios>
#include <iostream>
#include <istream>
#include <iterator>
#include <limits>
#include <list>
#include <map>
#include <memory>
#include <new>
#include <numeric>
#include <ostream>
#include <queue>
#include <random>
#include <set>
#include <sstream>
#include <stack>
#include <streambuf>
#include <string>
#include <tuple>
#include <type_traits>
#include <typeinfo>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
// utility
namespace Nyaan {
using ll = long long;
using i64 = long long;
using u64 = unsigned long long;
using i128 = __int128_t;
using u128 = __uint128_t;
template <typename T>
using V = vector<T>;
template <typename T>
using VV = vector<vector<T>>;
using vi = vector<int>;
using vl = vector<long long>;
using vd = V<double>;
using vs = V<string>;
using vvi = vector<vector<int>>;
using vvl = vector<vector<long long>>;
template <typename T>
using minpq = priority_queue<T, vector<T>, greater<T>>;
template <typename T, typename U>
struct P : pair<T, U> {
template <typename... Args>
P(Args... args) : pair<T, U>(args...) {}
using pair<T, U>::first;
using pair<T, U>::second;
P &operator+=(const P &r) {
first += r.first;
second += r.second;
return *this;
}
P &operator-=(const P &r) {
first -= r.first;
second -= r.second;
return *this;
}
P &operator*=(const P &r) {
first *= r.first;
second *= r.second;
return *this;
}
template <typename S>
P &operator*=(const S &r) {
first *= r, second *= r;
return *this;
}
P operator+(const P &r) const { return P(*this) += r; }
P operator-(const P &r) const { return P(*this) -= r; }
P operator*(const P &r) const { return P(*this) *= r; }
template <typename S>
P operator*(const S &r) const {
return P(*this) *= r;
}
P operator-() const { return P{-first, -second}; }
};
using pl = P<ll, ll>;
using pi = P<int, int>;
using vp = V<pl>;
constexpr int inf = 1001001001;
constexpr long long infLL = 4004004004004004004LL;
template <typename T>
int sz(const T &t) {
return t.size();
}
template <typename T, typename U>
inline bool amin(T &x, U y) {
return (y < x) ? (x = y, true) : false;
}
template <typename T, typename U>
inline bool amax(T &x, U y) {
return (x < y) ? (x = y, true) : false;
}
template <typename T>
inline T Max(const vector<T> &v) {
return *max_element(begin(v), end(v));
}
template <typename T>
inline T Min(const vector<T> &v) {
return *min_element(begin(v), end(v));
}
template <typename T>
inline long long Sum(const vector<T> &v) {
return accumulate(begin(v), end(v), 0LL);
}
template <typename T>
int lb(const vector<T> &v, const T &a) {
return lower_bound(begin(v), end(v), a) - begin(v);
}
template <typename T>
int ub(const vector<T> &v, const T &a) {
return upper_bound(begin(v), end(v), a) - begin(v);
}
constexpr long long TEN(int n) {
long long ret = 1, x = 10;
for (; n; x *= x, n >>= 1) ret *= (n & 1 ? x : 1);
return ret;
}
template <typename T, typename U>
pair<T, U> mkp(const T &t, const U &u) {
return make_pair(t, u);
}
template <typename T>
vector<T> mkrui(const vector<T> &v, bool rev = false) {
vector<T> ret(v.size() + 1);
if (rev) {
for (int i = int(v.size()) - 1; i >= 0; i--) ret[i] = v[i] + ret[i + 1];
} else {
for (int i = 0; i < int(v.size()); i++) ret[i + 1] = ret[i] + v[i];
}
return ret;
};
template <typename T>
vector<T> mkuni(const vector<T> &v) {
vector<T> ret(v);
sort(ret.begin(), ret.end());
ret.erase(unique(ret.begin(), ret.end()), ret.end());
return ret;
}
template <typename F>
vector<int> mkord(int N, F f) {
vector<int> ord(N);
iota(begin(ord), end(ord), 0);
sort(begin(ord), end(ord), f);
return ord;
}
template <typename T>
vector<int> mkinv(vector<T> &v) {
int max_val = *max_element(begin(v), end(v));
vector<int> inv(max_val + 1, -1);
for (int i = 0; i < (int)v.size(); i++) inv[v[i]] = i;
return inv;
}
vector<int> mkiota(int n) {
vector<int> ret(n);
iota(begin(ret), end(ret), 0);
return ret;
}
template <typename T>
T mkrev(const T &v) {
T w{v};
reverse(begin(w), end(w));
return w;
}
template <typename T>
bool nxp(vector<T> &v) {
return next_permutation(begin(v), end(v));
}
// 返り値の型は入力の T に依存
// i 要素目 : [0, a[i])
template <typename T>
vector<vector<T>> product(const vector<T> &a) {
vector<vector<T>> ret;
vector<T> v;
auto dfs = [&](auto rc, int i) -> void {
if (i == (int)a.size()) {
ret.push_back(v);
return;
}
for (int j = 0; j < a[i]; j++) v.push_back(j), rc(rc, i + 1), v.pop_back();
};
dfs(dfs, 0);
return ret;
}
// F : function(void(T&)), mod を取る操作
// T : 整数型のときはオーバーフローに注意する
template <typename T>
T Power(T a, long long n, const T &I, const function<void(T &)> &f) {
T res = I;
for (; n; f(a = a * a), n >>= 1) {
if (n & 1) f(res = res * a);
}
return res;
}
// T : 整数型のときはオーバーフローに注意する
template <typename T>
T Power(T a, long long n, const T &I = T{1}) {
return Power(a, n, I, function<void(T &)>{[](T &) -> void {}});
}
template <typename T>
T Rev(const T &v) {
T res = v;
reverse(begin(res), end(res));
return res;
}
template <typename T>
vector<T> Transpose(const vector<T> &v) {
using U = typename T::value_type;
int H = v.size(), W = v[0].size();
vector res(W, T(H, U{}));
for (int i = 0; i < H; i++) {
for (int j = 0; j < W; j++) {
res[j][i] = v[i][j];
}
}
return res;
}
template <typename T>
vector<T> Rotate(const vector<T> &v, int clockwise = true) {
using U = typename T::value_type;
int H = v.size(), W = v[0].size();
vector res(W, T(H, U{}));
for (int i = 0; i < H; i++) {
for (int j = 0; j < W; j++) {
if (clockwise) {
res[W - 1 - j][i] = v[i][j];
} else {
res[j][H - 1 - i] = v[i][j];
}
}
}
return res;
}
} // namespace Nyaan
// bit operation
namespace Nyaan {
__attribute__((target("popcnt"))) inline int popcnt(const u64 &a) {
return _mm_popcnt_u64(a);
}
inline int lsb(const u64 &a) { return a ? __builtin_ctzll(a) : 64; }
inline int ctz(const u64 &a) { return a ? __builtin_ctzll(a) : 64; }
inline int msb(const u64 &a) { return a ? 63 - __builtin_clzll(a) : -1; }
template <typename T>
inline int gbit(const T &a, int i) {
return (a >> i) & 1;
}
template <typename T>
inline void sbit(T &a, int i, bool b) {
if (gbit(a, i) != b) a ^= T(1) << i;
}
constexpr long long PW(int n) { return 1LL << n; }
constexpr long long MSK(int n) { return (1LL << n) - 1; }
} // namespace Nyaan
// inout
namespace Nyaan {
template <typename T, typename U>
ostream &operator<<(ostream &os, const pair<T, U> &p) {
os << p.first << " " << p.second;
return os;
}
template <typename T, typename U>
istream &operator>>(istream &is, pair<T, U> &p) {
is >> p.first >> p.second;
return is;
}
template <typename T>
ostream &operator<<(ostream &os, const vector<T> &v) {
int s = (int)v.size();
for (int i = 0; i < s; i++) os << (i ? " " : "") << v[i];
return os;
}
template <typename T>
istream &operator>>(istream &is, vector<T> &v) {
for (auto &x : v) is >> x;
return is;
}
istream &operator>>(istream &is, __int128_t &x) {
string S;
is >> S;
x = 0;
int flag = 0;
for (auto &c : S) {
if (c == '-') {
flag = true;
continue;
}
x *= 10;
x += c - '0';
}
if (flag) x = -x;
return is;
}
istream &operator>>(istream &is, __uint128_t &x) {
string S;
is >> S;
x = 0;
for (auto &c : S) {
x *= 10;
x += c - '0';
}
return is;
}
ostream &operator<<(ostream &os, __int128_t x) {
if (x == 0) return os << 0;
if (x < 0) os << '-', x = -x;
string S;
while (x) S.push_back('0' + x % 10), x /= 10;
reverse(begin(S), end(S));
return os << S;
}
ostream &operator<<(ostream &os, __uint128_t x) {
if (x == 0) return os << 0;
string S;
while (x) S.push_back('0' + x % 10), x /= 10;
reverse(begin(S), end(S));
return os << S;
}
void in() {}
template <typename T, class... U>
void in(T &t, U &...u) {
cin >> t;
in(u...);
}
void out() { cout << "\n"; }
template <typename T, class... U, char sep = ' '>
void out(const T &t, const U &...u) {
cout << t;
if (sizeof...(u)) cout << sep;
out(u...);
}
struct IoSetupNya {
IoSetupNya() {
cin.tie(nullptr);
ios::sync_with_stdio(false);
cout << fixed << setprecision(15);
cerr << fixed << setprecision(7);
}
} iosetupnya;
} // namespace Nyaan
// debug
#ifdef NyaanDebug
#define trc(...) (void(0))
#else
#define trc(...) (void(0))
#endif
#ifdef NyaanLocal
#define trc2(...) (void(0))
#else
#define trc2(...) (void(0))
#endif
// macro
#define each(x, v) for (auto&& x : v)
#define each2(x, y, v) for (auto&& [x, y] : v)
#define all(v) (v).begin(), (v).end()
#define rep(i, N) for (long long i = 0; i < (long long)(N); i++)
#define repr(i, N) for (long long i = (long long)(N)-1; i >= 0; i--)
#define rep1(i, N) for (long long i = 1; i <= (long long)(N); i++)
#define repr1(i, N) for (long long i = (N); (long long)(i) > 0; i--)
#define reg(i, a, b) for (long long i = (a); i < (b); i++)
#define regr(i, a, b) for (long long i = (b)-1; i >= (a); i--)
#define fi first
#define se second
#define ini(...) \
int __VA_ARGS__; \
in(__VA_ARGS__)
#define inl(...) \
long long __VA_ARGS__; \
in(__VA_ARGS__)
#define ins(...) \
string __VA_ARGS__; \
in(__VA_ARGS__)
#define in2(s, t) \
for (int i = 0; i < (int)s.size(); i++) { \
in(s[i], t[i]); \
}
#define in3(s, t, u) \
for (int i = 0; i < (int)s.size(); i++) { \
in(s[i], t[i], u[i]); \
}
#define in4(s, t, u, v) \
for (int i = 0; i < (int)s.size(); i++) { \
in(s[i], t[i], u[i], v[i]); \
}
#define die(...) \
do { \
Nyaan::out(__VA_ARGS__); \
return; \
} while (0)
namespace Nyaan {
void solve();
}
int main() { Nyaan::solve(); }
//
template <typename mint>
struct FormalPowerSeries : vector<mint> {
using vector<mint>::vector;
using FPS = FormalPowerSeries;
FPS &operator+=(const FPS &r) {
if (r.size() > this->size()) this->resize(r.size());
for (int i = 0; i < (int)r.size(); i++) (*this)[i] += r[i];
return *this;
}
FPS &operator+=(const mint &r) {
if (this->empty()) this->resize(1);
(*this)[0] += r;
return *this;
}
FPS &operator-=(const FPS &r) {
if (r.size() > this->size()) this->resize(r.size());
for (int i = 0; i < (int)r.size(); i++) (*this)[i] -= r[i];
return *this;
}
FPS &operator-=(const mint &r) {
if (this->empty()) this->resize(1);
(*this)[0] -= r;
return *this;
}
FPS &operator*=(const mint &v) {
for (int k = 0; k < (int)this->size(); k++) (*this)[k] *= v;
return *this;
}
FPS &operator/=(const FPS &r) {
if (this->size() < r.size()) {
this->clear();
return *this;
}
int n = this->size() - r.size() + 1;
if ((int)r.size() <= 64) {
FPS f(*this), g(r);
g.shrink();
mint coeff = g.back().inverse();
for (auto &x : g) x *= coeff;
int deg = (int)f.size() - (int)g.size() + 1;
int gs = g.size();
FPS quo(deg);
for (int i = deg - 1; i >= 0; i--) {
quo[i] = f[i + gs - 1];
for (int j = 0; j < gs; j++) f[i + j] -= quo[i] * g[j];
}
*this = quo * coeff;
this->resize(n, mint(0));
return *this;
}
return *this = ((*this).rev().pre(n) * r.rev().inv(n)).pre(n).rev();
}
FPS &operator%=(const FPS &r) {
*this -= *this / r * r;
shrink();
return *this;
}
FPS operator+(const FPS &r) const { return FPS(*this) += r; }
FPS operator+(const mint &v) const { return FPS(*this) += v; }
FPS operator-(const FPS &r) const { return FPS(*this) -= r; }
FPS operator-(const mint &v) const { return FPS(*this) -= v; }
FPS operator*(const FPS &r) const { return FPS(*this) *= r; }
FPS operator*(const mint &v) const { return FPS(*this) *= v; }
FPS operator/(const FPS &r) const { return FPS(*this) /= r; }
FPS operator%(const FPS &r) const { return FPS(*this) %= r; }
FPS operator-() const {
FPS ret(this->size());
for (int i = 0; i < (int)this->size(); i++) ret[i] = -(*this)[i];
return ret;
}
void shrink() {
while (this->size() && this->back() == mint(0)) this->pop_back();
}
FPS rev() const {
FPS ret(*this);
reverse(begin(ret), end(ret));
return ret;
}
FPS dot(FPS r) const {
FPS ret(min(this->size(), r.size()));
for (int i = 0; i < (int)ret.size(); i++) ret[i] = (*this)[i] * r[i];
return ret;
}
// 前 sz 項を取ってくる。sz に足りない項は 0 埋めする
FPS pre(int sz) const {
FPS ret(begin(*this), begin(*this) + min((int)this->size(), sz));
if ((int)ret.size() < sz) ret.resize(sz);
return ret;
}
FPS operator>>(int sz) const {
if ((int)this->size() <= sz) return {};
FPS ret(*this);
ret.erase(ret.begin(), ret.begin() + sz);
return ret;
}
FPS operator<<(int sz) const {
FPS ret(*this);
ret.insert(ret.begin(), sz, mint(0));
return ret;
}
FPS diff() const {
const int n = (int)this->size();
FPS ret(max(0, n - 1));
mint one(1), coeff(1);
for (int i = 1; i < n; i++) {
ret[i - 1] = (*this)[i] * coeff;
coeff += one;
}
return ret;
}
FPS integral() const {
const int n = (int)this->size();
FPS ret(n + 1);
ret[0] = mint(0);
if (n > 0) ret[1] = mint(1);
auto mod = mint::get_mod();
for (int i = 2; i <= n; i++) ret[i] = (-ret[mod % i]) * (mod / i);
for (int i = 0; i < n; i++) ret[i + 1] *= (*this)[i];
return ret;
}
mint eval(mint x) const {
mint r = 0, w = 1;
for (auto &v : *this) r += w * v, w *= x;
return r;
}
FPS log(int deg = -1) const {
assert(!(*this).empty() && (*this)[0] == mint(1));
if (deg == -1) deg = (int)this->size();
return (this->diff() * this->inv(deg)).pre(deg - 1).integral();
}
FPS pow(int64_t k, int deg = -1) const {
const int n = (int)this->size();
if (deg == -1) deg = n;
if (k == 0) {
FPS ret(deg);
if (deg) ret[0] = 1;
return ret;
}
for (int i = 0; i < n; i++) {
if ((*this)[i] != mint(0)) {
mint rev = mint(1) / (*this)[i];
FPS ret = (((*this * rev) >> i).log(deg) * k).exp(deg);
ret *= (*this)[i].pow(k);
ret = (ret << (i * k)).pre(deg);
if ((int)ret.size() < deg) ret.resize(deg, mint(0));
return ret;
}
if (__int128_t(i + 1) * k >= deg) return FPS(deg, mint(0));
}
return FPS(deg, mint(0));
}
static void *ntt_ptr;
static void set_fft();
FPS &operator*=(const FPS &r);
void ntt();
void intt();
void ntt_doubling();
static int ntt_pr();
FPS inv(int deg = -1) const;
FPS exp(int deg = -1) const;
};
template <typename mint>
void *FormalPowerSeries<mint>::ntt_ptr = nullptr;
/**
* @brief 多項式/形式的冪級数ライブラリ
* @docs docs/fps/formal-power-series.md
*/
namespace poly_gcd {
template <typename mint>
using FPS = FormalPowerSeries<mint>;
template <typename mint>
using Arr = pair<FPS<mint>, FPS<mint>>;
template <typename mint>
struct Mat {
using fps = FPS<mint>;
fps a00, a01, a10, a11;
Mat() = default;
Mat(const fps& a00_, const fps& a01_, const fps& a10_, const fps& a11_)
: a00(a00_), a01(a01_), a10(a10_), a11(a11_) {}
Mat& operator*=(const Mat& r) {
fps A00 = a00 * r.a00 + a01 * r.a10;
fps A01 = a00 * r.a01 + a01 * r.a11;
fps A10 = a10 * r.a00 + a11 * r.a10;
fps A11 = a10 * r.a01 + a11 * r.a11;
A00.shrink();
A01.shrink();
A10.shrink();
A11.shrink();
swap(A00, a00);
swap(A01, a01);
swap(A10, a10);
swap(A11, a11);
return *this;
}
static Mat I() { return Mat(fps{mint(1)}, fps(), fps(), fps{mint(1)}); }
Mat operator*(const Mat& r) const { return Mat(*this) *= r; }
};
template <typename mint>
Arr<mint> operator*(const Mat<mint>& m, const Arr<mint>& a) {
using fps = FPS<mint>;
fps b0 = m.a00 * a.first + m.a01 * a.second;
fps b1 = m.a10 * a.first + m.a11 * a.second;
b0.shrink();
b1.shrink();
return {b0, b1};
};
template <typename mint>
void InnerNaiveGCD(Mat<mint>& m, Arr<mint>& p) {
using fps = FPS<mint>;
fps quo = p.first / p.second;
fps rem = p.first - p.second * quo;
fps b10 = m.a00 - m.a10 * quo;
fps b11 = m.a01 - m.a11 * quo;
rem.shrink();
b10.shrink();
b11.shrink();
swap(b10, m.a10);
swap(b11, m.a11);
swap(b10, m.a00);
swap(b11, m.a01);
p = {p.second, rem};
}
template <typename mint>
Mat<mint> InnerHalfGCD(Arr<mint> p) {
int n = p.first.size(), m = p.second.size();
int k = (n + 1) / 2;
if (m <= k) return Mat<mint>::I();
Mat<mint> m1 = InnerHalfGCD(make_pair(p.first >> k, p.second >> k));
p = m1 * p;
if ((int)p.second.size() <= k) return m1;
InnerNaiveGCD(m1, p);
if ((int)p.second.size() <= k) return m1;
int l = (int)p.first.size() - 1;
int j = 2 * k - l;
p.first = p.first >> j;
p.second = p.second >> j;
return InnerHalfGCD(p) * m1;
}
template <typename mint>
Mat<mint> InnerPolyGCD(const FPS<mint>& a, const FPS<mint>& b) {
Arr<mint> p{a, b};
p.first.shrink();
p.second.shrink();
int n = p.first.size(), m = p.second.size();
if (n < m) {
Mat<mint> mat = InnerPolyGCD(p.second, p.first);
swap(mat.a00, mat.a01);
swap(mat.a10, mat.a11);
return mat;
}
Mat<mint> res = Mat<mint>::I();
while (1) {
Mat<mint> m1 = InnerHalfGCD(p);
p = m1 * p;
if (p.second.empty()) return m1 * res;
InnerNaiveGCD(m1, p);
if (p.second.empty()) return m1 * res;
res = m1 * res;
}
}
template <typename mint>
FPS<mint> PolyGCD(const FPS<mint>& a, const FPS<mint>& b) {
Arr<mint> p(a, b);
Mat<mint> m = InnerPolyGCD(a, b);
p = m * p;
if (!p.first.empty()) {
mint coeff = p.first.back().inverse();
for (auto& x : p.first) x *= coeff;
}
return p.first;
}
template <typename mint>
pair<int, FPS<mint>> PolyInv(const FPS<mint>& f, const FPS<mint>& g) {
using fps = FPS<mint>;
pair<fps, fps> p(f, g);
Mat<mint> m = InnerPolyGCD(f, g);
fps gcd_ = (m * p).first;
if (gcd_.size() != 1) return {false, fps()};
pair<fps, fps> x(fps{mint(1)}, g);
return {true, ((m * x).first % g) * gcd_[0].inverse()};
}
} // namespace poly_gcd
using poly_gcd::PolyGCD;
using poly_gcd::PolyInv;
/**
* @brief 多項式GCD
* @docs docs/fps/polynomial-gcd.md
*/
//
template <typename T>
struct edge {
int src, to;
T cost;
edge(int _to, T _cost) : src(-1), to(_to), cost(_cost) {}
edge(int _src, int _to, T _cost) : src(_src), to(_to), cost(_cost) {}
edge &operator=(const int &x) {
to = x;
return *this;
}
operator int() const { return to; }
};
template <typename T>
using Edges = vector<edge<T>>;
template <typename T>
using WeightedGraph = vector<Edges<T>>;
using UnweightedGraph = vector<vector<int>>;
// Input of (Unweighted) Graph
UnweightedGraph graph(int N, int M = -1, bool is_directed = false,
bool is_1origin = true) {
UnweightedGraph g(N);
if (M == -1) M = N - 1;
for (int _ = 0; _ < M; _++) {
int x, y;
cin >> x >> y;
if (is_1origin) x--, y--;
g[x].push_back(y);
if (!is_directed) g[y].push_back(x);
}
return g;
}
// Input of Weighted Graph
template <typename T>
WeightedGraph<T> wgraph(int N, int M = -1, bool is_directed = false,
bool is_1origin = true) {
WeightedGraph<T> g(N);
if (M == -1) M = N - 1;
for (int _ = 0; _ < M; _++) {
int x, y;
cin >> x >> y;
T c;
cin >> c;
if (is_1origin) x--, y--;
g[x].emplace_back(x, y, c);
if (!is_directed) g[y].emplace_back(y, x, c);
}
return g;
}
// Input of Edges
template <typename T>
Edges<T> esgraph(int N, int M, int is_weighted = true, bool is_1origin = true) {
Edges<T> es;
for (int _ = 0; _ < M; _++) {
int x, y;
cin >> x >> y;
T c;
if (is_weighted)
cin >> c;
else
c = 1;
if (is_1origin) x--, y--;
es.emplace_back(x, y, c);
}
return es;
}
// Input of Adjacency Matrix
template <typename T>
vector<vector<T>> adjgraph(int N, int M, T INF, int is_weighted = true,
bool is_directed = false, bool is_1origin = true) {
vector<vector<T>> d(N, vector<T>(N, INF));
for (int _ = 0; _ < M; _++) {
int x, y;
cin >> x >> y;
T c;
if (is_weighted)
cin >> c;
else
c = 1;
if (is_1origin) x--, y--;
d[x][y] = c;
if (!is_directed) d[y][x] = c;
}
return d;
}
/**
* @brief グラフテンプレート
* @docs docs/graph/graph-template.md
*/
// 一般のグラフのstからの距離！！！！
// unvisited nodes : d = -1
vector<int> Depth(const UnweightedGraph &g, int start = 0) {
int n = g.size();
vector<int> ds(n, -1);
ds[start] = 0;
queue<int> q;
q.push(start);
while (!q.empty()) {
int c = q.front();
q.pop();
int dc = ds[c];
for (auto &d : g[c]) {
if (ds[d] == -1) {
ds[d] = dc + 1;
q.push(d);
}
}
}
return ds;
}
// Depth of Rooted Weighted Tree
// unvisited nodes : d = -1
template <typename T>
vector<T> Depth(const WeightedGraph<T> &g, int start = 0) {
vector<T> d(g.size(), -1);
auto dfs = [&](auto rec, int cur, T val, int par = -1) -> void {
d[cur] = val;
for (auto &dst : g[cur]) {
if (dst == par) continue;
rec(rec, dst, val + dst.cost, cur);
}
};
dfs(dfs, start, 0);
return d;
}
// Diameter of Tree
// return value : { {u, v}, length }
pair<pair<int, int>, int> Diameter(const UnweightedGraph &g) {
auto d = Depth(g, 0);
int u = max_element(begin(d), end(d)) - begin(d);
d = Depth(g, u);
int v = max_element(begin(d), end(d)) - begin(d);
return make_pair(make_pair(u, v), d[v]);
}
// Diameter of Weighted Tree
// return value : { {u, v}, length }
template <typename T>
pair<pair<int, int>, T> Diameter(const WeightedGraph<T> &g) {
auto d = Depth(g, 0);
int u = max_element(begin(d), end(d)) - begin(d);
d = Depth(g, u);
int v = max_element(begin(d), end(d)) - begin(d);
return make_pair(make_pair(u, v), d[v]);
}
// nodes on the path u-v ( O(N) )
template <typename G>
vector<int> Path(G &g, int u, int v) {
vector<int> ret;
int end = 0;
auto dfs = [&](auto rec, int cur, int par = -1) -> void {
ret.push_back(cur);
if (cur == v) {
end = 1;
return;
}
for (int dst : g[cur]) {
if (dst == par) continue;
rec(rec, dst, cur);
if (end) return;
}
if (end) return;
ret.pop_back();
};
dfs(dfs, u);
return ret;
}
/**
* @brief グラフユーティリティ
* @docs docs/graph/graph-utility.md
*/
template <typename T>
struct has_cost {
private:
template <typename U>
static auto confirm(U u) -> decltype(u.cost, std::true_type());
static auto confirm(...) -> std::false_type;
public:
enum : bool { value = decltype(confirm(std::declval<T>()))::value };
};
template <typename T>
vector<vector<T>> inverse_tree(const vector<vector<T>>& g) {
int N = (int)g.size();
vector<vector<T>> rg(N);
for (int i = 0; i < N; i++) {
for (auto& e : g[i]) {
if constexpr (is_same<T, int>::value) {
rg[e].push_back(i);
} else if constexpr (has_cost<T>::value) {
rg[e].emplace_back(e.to, i, e.cost);
} else {
assert(0);
}
}
}
return rg;
}
template <typename T>
vector<vector<T>> rooted_tree(const vector<vector<T>>& g, int root = 0) {
int N = (int)g.size();
vector<vector<T>> rg(N);
vector<char> v(N, false);
v[root] = true;
queue<int> que;
que.emplace(root);
while (!que.empty()) {
auto p = que.front();
que.pop();
for (auto& e : g[p]) {
if (v[e] == false) {
v[e] = true;
que.push(e);
rg[p].push_back(e);
}
}
}
return rg;
}
/**
* @brief 根付き木・逆辺からなる木への変換
*/
template <typename G>
struct HeavyLightDecomposition {
private:
void dfs_sz(int cur) {
size[cur] = 1;
for (auto& dst : g[cur]) {
if (dst == par[cur]) {
if (g[cur].size() >= 2 && int(dst) == int(g[cur][0]))
swap(g[cur][0], g[cur][1]);
else
continue;
}
depth[dst] = depth[cur] + 1;
par[dst] = cur;
dfs_sz(dst);
size[cur] += size[dst];
if (size[dst] > size[g[cur][0]]) {
swap(dst, g[cur][0]);
}
}
}
void dfs_hld(int cur) {
down[cur] = id++;
for (auto dst : g[cur]) {
if (dst == par[cur]) continue;
nxt[dst] = (int(dst) == int(g[cur][0]) ? nxt[cur] : int(dst));
dfs_hld(dst);
}
up[cur] = id;
}
// [u, v)
vector<pair<int, int>> ascend(int u, int v) const {
vector<pair<int, int>> res;
while (nxt[u] != nxt[v]) {
res.emplace_back(down[u], down[nxt[u]]);
u = par[nxt[u]];
}
if (u != v) res.emplace_back(down[u], down[v] + 1);
return res;
}
// (u, v]
vector<pair<int, int>> descend(int u, int v) const {
if (u == v) return {};
if (nxt[u] == nxt[v]) return {{down[u] + 1, down[v]}};
auto res = descend(u, par[nxt[v]]);
res.emplace_back(down[nxt[v]], down[v]);
return res;
}
public:
G& g;
int id;
vector<int> size, depth, down, up, nxt, par;
HeavyLightDecomposition(G& _g, int root = 0)
: g(_g),
id(0),
size(g.size(), 0),
depth(g.size(), 0),
down(g.size(), -1),
up(g.size(), -1),
nxt(g.size(), root),
par(g.size(), root) {
dfs_sz(root);
dfs_hld(root);
}
void build(int root) {
dfs_sz(root);
dfs_hld(root);
}
pair<int, int> idx(int i) const { return make_pair(down[i], up[i]); }
template <typename F>
void path_query(int u, int v, bool vertex, const F& f) {
int l = lca(u, v);
for (auto&& [a, b] : ascend(u, l)) {
int s = a + 1, t = b;
s > t ? f(t, s) : f(s, t);
}
if (vertex) f(down[l], down[l] + 1);
for (auto&& [a, b] : descend(l, v)) {
int s = a, t = b + 1;
s > t ? f(t, s) : f(s, t);
}
}
template <typename F>
void path_noncommutative_query(int u, int v, bool vertex, const F& f) {
int l = lca(u, v);
for (auto&& [a, b] : ascend(u, l)) f(a + 1, b);
if (vertex) f(down[l], down[l] + 1);
for (auto&& [a, b] : descend(l, v)) f(a, b + 1);
}
template <typename F>
void subtree_query(int u, bool vertex, const F& f) {
f(down[u] + int(!vertex), up[u]);
}
int lca(int a, int b) {
while (nxt[a] != nxt[b]) {
if (down[a] < down[b]) swap(a, b);
a = par[nxt[a]];
}
return depth[a] < depth[b] ? a : b;
}
int dist(int a, int b) { return depth[a] + depth[b] - depth[lca(a, b)] * 2; }
};
/**
* @brief Heavy Light Decomposition(重軽分解)
* @docs docs/tree/heavy-light-decomposition.md
*/
//
template <typename fps>
struct fps_fraction {
using frac = fps_fraction;
using mint = typename fps::value_type;
fps p, q;
fps_fraction(const fps& numerator = fps{0}, const fps& denominator = fps{1})
: p(numerator), q(denominator) {}
friend frac operator+(const frac& l, const frac& r) {
return frac{l.p * r.q + r.p * l.q, l.q * r.q};
}
friend frac operator-(const frac& l, const frac& r) {
return frac{l.p * r.q - r.p * l.q, l.q * r.q};
}
friend frac operator*(const frac& l, const frac& r) {
return frac{l.p * r.p, l.q * r.q};
}
friend frac operator/(const frac& l, const frac& r) {
return frac{l.p * r.q, l.q * r.p};
}
frac& operator+=(const mint& r) {
(*this).p += (*this).q * r;
return *this;
}
frac& operator-=(const mint& r) {
(*this).p -= (*this).q * r;
return *this;
}
frac& operator*=(const mint& r) {
(*this).p *= r;
return *this;
}
frac operator+(const mint& r) { return frac{*this} += r; }
frac operator-(const mint& r) { return frac{*this} -= r; }
frac operator*(const mint& r) { return frac{*this} *= r; }
frac operator/(const mint& r) { return frac{*this} /= r; }
frac& operator+=(const frac& r) { return *this = (*this) + r; }
frac& operator-=(const frac& r) { return *this = (*this) - r; }
frac& operator*=(const frac& r) { return *this = (*this) * r; }
frac operator-() const { return frac{-p, q}; }
frac inverse() const { return frac{q, p}; };
void shrink() { p.shrink(), q.shrink(); }
friend bool operator==(const frac& l, const frac& r) {
return l.p == r.p && l.q == r.q;
}
friend bool operator!=(const frac& l, const frac& r) {
return l.p != r.p || l.q != r.q;
}
friend ostream& operator<<(ostream& os, const frac& r) {
os << "[ ";
for (auto& x : r.p) os << x << ", ";
os << "], ";
os << "[ ";
for (auto& x : r.q) os << x << ", ";
os << " ]";
return os;
}
};
template <typename mint>
mint LinearRecurrence(long long k, FormalPowerSeries<mint> Q,
FormalPowerSeries<mint> P) {
Q.shrink();
mint ret = 0;
if (P.size() >= Q.size()) {
auto R = P / Q;
P -= R * Q;
P.shrink();
if (k < (int)R.size()) ret += R[k];
}
if ((int)P.size() == 0) return ret;
FormalPowerSeries<mint>::set_fft();
if (FormalPowerSeries<mint>::ntt_ptr == nullptr) {
P.resize((int)Q.size() - 1);
while (k) {
auto Q2 = Q;
for (int i = 1; i < (int)Q2.size(); i += 2) Q2[i] = -Q2[i];
auto S = P * Q2;
auto T = Q * Q2;
if (k & 1) {
for (int i = 1; i < (int)S.size(); i += 2) P[i >> 1] = S[i];
for (int i = 0; i < (int)T.size(); i += 2) Q[i >> 1] = T[i];
} else {
for (int i = 0; i < (int)S.size(); i += 2) P[i >> 1] = S[i];
for (int i = 0; i < (int)T.size(); i += 2) Q[i >> 1] = T[i];
}
k >>= 1;
}
return ret + P[0];
} else {
int N = 1;
while (N < (int)Q.size()) N <<= 1;
P.resize(2 * N);
Q.resize(2 * N);
P.ntt();
Q.ntt();
vector<mint> S(2 * N), T(2 * N);
vector<int> btr(N);
for (int i = 0, logn = __builtin_ctz(N); i < (1 << logn); i++) {
btr[i] = (btr[i >> 1] >> 1) + ((i & 1) << (logn - 1));
}
mint dw = mint(FormalPowerSeries<mint>::ntt_pr())
.inverse()
.pow((mint::get_mod() - 1) / (2 * N));
while (k) {
mint inv2 = mint(2).inverse();
// even degree of Q(x)Q(-x)
T.resize(N);
for (int i = 0; i < N; i++) T[i] = Q[(i << 1) | 0] * Q[(i << 1) | 1];
S.resize(N);
if (k & 1) {
// odd degree of P(x)Q(-x)
for (auto &i : btr) {
S[i] = (P[(i << 1) | 0] * Q[(i << 1) | 1] -
P[(i << 1) | 1] * Q[(i << 1) | 0]) *
inv2;
inv2 *= dw;
}
} else {
// even degree of P(x)Q(-x)
for (int i = 0; i < N; i++) {
S[i] = (P[(i << 1) | 0] * Q[(i << 1) | 1] +
P[(i << 1) | 1] * Q[(i << 1) | 0]) *
inv2;
}
}
swap(P, S);
swap(Q, T);
k >>= 1;
if (k < N) break;
P.ntt_doubling();
Q.ntt_doubling();
}
P.intt();
Q.intt();
return ret + (P * (Q.inv()))[k];
}
}
template <typename mint>
mint kitamasa(long long N, FormalPowerSeries<mint> Q,
FormalPowerSeries<mint> a) {
assert(!Q.empty() && Q[0] != 0);
if (N < (int)a.size()) return a[N];
assert((int)a.size() >= int(Q.size()) - 1);
auto P = a.pre((int)Q.size() - 1) * Q;
P.resize(Q.size() - 1);
return LinearRecurrence<mint>(N, Q, P);
}
/**
* @brief 線形漸化式の高速計算
* @docs docs/fps/kitamasa.md
*/
using namespace std;
// {rank, det(非正方行列の場合は未定義)} を返す
// 型が double や Rational でも動くはず？(未検証)
//
// pivot 候補 : [0, pivot_end)
template <typename T>
std::pair<int, T> GaussElimination(vector<vector<T>> &a, int pivot_end = -1,
bool diagonalize = false) {
int H = a.size(), W = a[0].size(), rank = 0;
if (pivot_end == -1) pivot_end = W;
T det = 1;
for (int j = 0; j < pivot_end; j++) {
int idx = -1;
for (int i = rank; i < H; i++) {
if (a[i][j] != T(0)) {
idx = i;
break;
}
}
if (idx == -1) {
det = 0;
continue;
}
if (rank != idx) det = -det, swap(a[rank], a[idx]);
det *= a[rank][j];
if (diagonalize && a[rank][j] != T(1)) {
T coeff = T(1) / a[rank][j];
for (int k = j; k < W; k++) a[rank][k] *= coeff;
}
int is = diagonalize ? 0 : rank + 1;
for (int i = is; i < H; i++) {
if (i == rank) continue;
if (a[i][j] != T(0)) {
T coeff = a[i][j] / a[rank][j];
for (int k = j; k < W; k++) a[i][k] -= a[rank][k] * coeff;
}
}
rank++;
}
return make_pair(rank, det);
}
template <typename mint>
vector<vector<mint>> inverse_matrix(const vector<vector<mint>>& a) {
int N = a.size();
assert(N > 0);
assert(N == (int)a[0].size());
vector<vector<mint>> m(N, vector<mint>(2 * N));
for (int i = 0; i < N; i++) {
copy(begin(a[i]), end(a[i]), begin(m[i]));
m[i][N + i] = 1;
}
auto [rank, det] = GaussElimination(m, N, true);
if (rank != N) return {};
vector<vector<mint>> b(N);
for (int i = 0; i < N; i++) {
copy(begin(m[i]) + N, end(m[i]), back_inserter(b[i]));
}
return b;
}
template <class T>
struct Matrix {
vector<vector<T> > A;
Matrix() = default;
Matrix(int n, int m) : A(n, vector<T>(m, T())) {}
Matrix(int n) : A(n, vector<T>(n, T())){};
int H() const { return A.size(); }
int W() const { return A[0].size(); }
int size() const { return A.size(); }
inline const vector<T> &operator[](int k) const { return A[k]; }
inline vector<T> &operator[](int k) { return A[k]; }
static Matrix I(int n) {
Matrix mat(n);
for (int i = 0; i < n; i++) mat[i][i] = 1;
return (mat);
}
Matrix &operator+=(const Matrix &B) {
int n = H(), m = W();
assert(n == B.H() && m == B.W());
for (int i = 0; i < n; i++)
for (int j = 0; j < m; j++) (*this)[i][j] += B[i][j];
return (*this);
}
Matrix &operator-=(const Matrix &B) {
int n = H(), m = W();
assert(n == B.H() && m == B.W());
for (int i = 0; i < n; i++)
for (int j = 0; j < m; j++) (*this)[i][j] -= B[i][j];
return (*this);
}
Matrix &operator*=(const Matrix &B) {
int n = H(), m = B.W(), p = W();
assert(p == B.H());
vector<vector<T> > C(n, vector<T>(m, T{}));
for (int i = 0; i < n; i++)
for (int k = 0; k < p; k++)
for (int j = 0; j < m; j++) C[i][j] += (*this)[i][k] * B[k][j];
A.swap(C);
return (*this);
}
Matrix &operator^=(long long k) {
Matrix B = Matrix::I(H());
while (k > 0) {
if (k & 1) B *= *this;
*this *= *this;
k >>= 1LL;
}
A.swap(B.A);
return (*this);
}
Matrix operator+(const Matrix &B) const { return (Matrix(*this) += B); }
Matrix operator-(const Matrix &B) const { return (Matrix(*this) -= B); }
Matrix operator*(const Matrix &B) const { return (Matrix(*this) *= B); }
Matrix operator^(const long long k) const { return (Matrix(*this) ^= k); }
bool operator==(const Matrix &B) const {
assert(H() == B.H() && W() == B.W());
for (int i = 0; i < H(); i++)
for (int j = 0; j < W(); j++)
if (A[i][j] != B[i][j]) return false;
return true;
}
bool operator!=(const Matrix &B) const {
assert(H() == B.H() && W() == B.W());
for (int i = 0; i < H(); i++)
for (int j = 0; j < W(); j++)
if (A[i][j] != B[i][j]) return true;
return false;
}
Matrix inverse() const {
assert(H() == W());
Matrix B(H());
B.A = inverse_matrix(A);
return B;
}
friend ostream &operator<<(ostream &os, const Matrix &p) {
int n = p.H(), m = p.W();
for (int i = 0; i < n; i++) {
os << (i ? " " : "") << "[";
for (int j = 0; j < m; j++) {
os << p[i][j] << (j + 1 == m ? "]\n" : ",");
}
}
return (os);
}
T determinant() const {
Matrix B(*this);
assert(H() == W());
T ret = 1;
for (int i = 0; i < H(); i++) {
int idx = -1;
for (int j = i; j < W(); j++) {
if (B[j][i] != 0) {
idx = j;
break;
}
}
if (idx == -1) return 0;
if (i != idx) {
ret *= T(-1);
swap(B[i], B[idx]);
}
ret *= B[i][i];
T inv = T(1) / B[i][i];
for (int j = 0; j < W(); j++) {
B[i][j] *= inv;
}
for (int j = i + 1; j < H(); j++) {
T a = B[j][i];
if (a == 0) continue;
for (int k = i; k < W(); k++) {
B[j][k] -= B[i][k] * a;
}
}
}
return ret;
}
};
/**
* @brief 行列ライブラリ
*/
//
template <typename mint>
struct NTT {
static constexpr uint32_t get_pr() {
uint32_t _mod = mint::get_mod();
using u64 = uint64_t;
u64 ds[32] = {};
int idx = 0;
u64 m = _mod - 1;
for (u64 i = 2; i * i <= m; ++i) {
if (m % i == 0) {
ds[idx++] = i;
while (m % i == 0) m /= i;
}
}
if (m != 1) ds[idx++] = m;
uint32_t _pr = 2;
while (1) {
int flg = 1;
for (int i = 0; i < idx; ++i) {
u64 a = _pr, b = (_mod - 1) / ds[i], r = 1;
while (b) {
if (b & 1) r = r * a % _mod;
a = a * a % _mod;
b >>= 1;
}
if (r == 1) {
flg = 0;
break;
}
}
if (flg == 1) break;
++_pr;
}
return _pr;
};
static constexpr uint32_t mod = mint::get_mod();
static constexpr uint32_t pr = get_pr();
static constexpr int level = __builtin_ctzll(mod - 1);
mint dw[level], dy[level];
void setwy(int k) {
mint w[level], y[level];
w[k - 1] = mint(pr).pow((mod - 1) / (1 << k));
y[k - 1] = w[k - 1].inverse();
for (int i = k - 2; i > 0; --i)
w[i] = w[i + 1] * w[i + 1], y[i] = y[i + 1] * y[i + 1];
dw[1] = w[1], dy[1] = y[1], dw[2] = w[2], dy[2] = y[2];
for (int i = 3; i < k; ++i) {
dw[i] = dw[i - 1] * y[i - 2] * w[i];
dy[i] = dy[i - 1] * w[i - 2] * y[i];
}
}
NTT() { setwy(level); }
void fft4(vector<mint> &a, int k) {
if ((int)a.size() <= 1) return;
if (k == 1) {
mint a1 = a[1];
a[1] = a[0] - a[1];
a[0] = a[0] + a1;
return;
}
if (k & 1) {
int v = 1 << (k - 1);
for (int j = 0; j < v; ++j) {
mint ajv = a[j + v];
a[j + v] = a[j] - ajv;
a[j] += ajv;
}
}
int u = 1 << (2 + (k & 1));
int v = 1 << (k - 2 - (k & 1));
mint one = mint(1);
mint imag = dw[1];
while (v) {
// jh = 0
{
int j0 = 0;
int j1 = v;
int j2 = j1 + v;
int j3 = j2 + v;
for (; j0 < v; ++j0, ++j1, ++j2, ++j3) {
mint t0 = a[j0], t1 = a[j1], t2 = a[j2], t3 = a[j3];
mint t0p2 = t0 + t2, t1p3 = t1 + t3;
mint t0m2 = t0 - t2, t1m3 = (t1 - t3) * imag;
a[j0] = t0p2 + t1p3, a[j1] = t0p2 - t1p3;
a[j2] = t0m2 + t1m3, a[j3] = t0m2 - t1m3;
}
}
// jh >= 1
mint ww = one, xx = one * dw[2], wx = one;
for (int jh = 4; jh < u;) {
ww = xx * xx, wx = ww * xx;
int j0 = jh * v;
int je = j0 + v;
int j2 = je + v;
for (; j0 < je; ++j0, ++j2) {
mint t0 = a[j0], t1 = a[j0 + v] * xx, t2 = a[j2] * ww,
t3 = a[j2 + v] * wx;
mint t0p2 = t0 + t2, t1p3 = t1 + t3;
mint t0m2 = t0 - t2, t1m3 = (t1 - t3) * imag;
a[j0] = t0p2 + t1p3, a[j0 + v] = t0p2 - t1p3;
a[j2] = t0m2 + t1m3, a[j2 + v] = t0m2 - t1m3;
}
xx *= dw[__builtin_ctzll((jh += 4))];
}
u <<= 2;
v >>= 2;
}
}
void ifft4(vector<mint> &a, int k) {
if ((int)a.size() <= 1) return;
if (k == 1) {
mint a1 = a[1];
a[1] = a[0] - a[1];
a[0] = a[0] + a1;
return;
}
int u = 1 << (k - 2);
int v = 1;
mint one = mint(1);
mint imag = dy[1];
while (u) {
// jh = 0
{
int j0 = 0;
int j1 = v;
int j2 = v + v;
int j3 = j2 + v;
for (; j0 < v; ++j0, ++j1, ++j2, ++j3) {
mint t0 = a[j0], t1 = a[j1], t2 = a[j2], t3 = a[j3];
mint t0p1 = t0 + t1, t2p3 = t2 + t3;
mint t0m1 = t0 - t1, t2m3 = (t2 - t3) * imag;
a[j0] = t0p1 + t2p3, a[j2] = t0p1 - t2p3;
a[j1] = t0m1 + t2m3, a[j3] = t0m1 - t2m3;
}
}
// jh >= 1
mint ww = one, xx = one * dy[2], yy = one;
u <<= 2;
for (int jh = 4; jh < u;) {
ww = xx * xx, yy = xx * imag;
int j0 = jh * v;
int je = j0 + v;
int j2 = je + v;
for (; j0 < je; ++j0, ++j2) {
mint t0 = a[j0], t1 = a[j0 + v], t2 = a[j2], t3 = a[j2 + v];
mint t0p1 = t0 + t1, t2p3 = t2 + t3;
mint t0m1 = (t0 - t1) * xx, t2m3 = (t2 - t3) * yy;
a[j0] = t0p1 + t2p3, a[j2] = (t0p1 - t2p3) * ww;
a[j0 + v] = t0m1 + t2m3, a[j2 + v] = (t0m1 - t2m3) * ww;
}
xx *= dy[__builtin_ctzll(jh += 4)];
}
u >>= 4;
v <<= 2;
}
if (k & 1) {
u = 1 << (k - 1);
for (int j = 0; j < u; ++j) {
mint ajv = a[j] - a[j + u];
a[j] += a[j + u];
a[j + u] = ajv;
}
}
}
void ntt(vector<mint> &a) {
if ((int)a.size() <= 1) return;
fft4(a, __builtin_ctz(a.size()));
}
void intt(vector<mint> &a) {
if ((int)a.size() <= 1) return;
ifft4(a, __builtin_ctz(a.size()));
mint iv = mint(a.size()).inverse();
for (auto &x : a) x *= iv;
}
vector<mint> multiply(const vector<mint> &a, const vector<mint> &b) {
int l = a.size() + b.size() - 1;
if (min<int>(a.size(), b.size()) <= 40) {
vector<mint> s(l);
for (int i = 0; i < (int)a.size(); ++i)
for (int j = 0; j < (int)b.size(); ++j) s[i + j] += a[i] * b[j];
return s;
}
int k = 2, M = 4;
while (M < l) M <<= 1, ++k;
setwy(k);
vector<mint> s(M);
for (int i = 0; i < (int)a.size(); ++i) s[i] = a[i];
fft4(s, k);
if (a.size() == b.size() && a == b) {
for (int i = 0; i < M; ++i) s[i] *= s[i];
} else {
vector<mint> t(M);
for (int i = 0; i < (int)b.size(); ++i) t[i] = b[i];
fft4(t, k);
for (int i = 0; i < M; ++i) s[i] *= t[i];
}
ifft4(s, k);
s.resize(l);
mint invm = mint(M).inverse();
for (int i = 0; i < l; ++i) s[i] *= invm;
return s;
}
void ntt_doubling(vector<mint> &a) {
int M = (int)a.size();
auto b = a;
intt(b);
mint r = 1, zeta = mint(pr).pow((mint::get_mod() - 1) / (M << 1));
for (int i = 0; i < M; i++) b[i] *= r, r *= zeta;
ntt(b);
copy(begin(b), end(b), back_inserter(a));
}
};
template <typename mint>
void FormalPowerSeries<mint>::set_fft() {
if (!ntt_ptr) ntt_ptr = new NTT<mint>;
}
template <typename mint>
FormalPowerSeries<mint>& FormalPowerSeries<mint>::operator*=(
const FormalPowerSeries<mint>& r) {
if (this->empty() || r.empty()) {
this->clear();
return *this;
}
set_fft();
auto ret = static_cast<NTT<mint>*>(ntt_ptr)->multiply(*this, r);
return *this = FormalPowerSeries<mint>(ret.begin(), ret.end());
}
template <typename mint>
void FormalPowerSeries<mint>::ntt() {
set_fft();
static_cast<NTT<mint>*>(ntt_ptr)->ntt(*this);
}
template <typename mint>
void FormalPowerSeries<mint>::intt() {
set_fft();
static_cast<NTT<mint>*>(ntt_ptr)->intt(*this);
}
template <typename mint>
void FormalPowerSeries<mint>::ntt_doubling() {
set_fft();
static_cast<NTT<mint>*>(ntt_ptr)->ntt_doubling(*this);
}
template <typename mint>
int FormalPowerSeries<mint>::ntt_pr() {
set_fft();
return static_cast<NTT<mint>*>(ntt_ptr)->pr;
}
template <typename mint>
FormalPowerSeries<mint> FormalPowerSeries<mint>::inv(int deg) const {
assert((*this)[0] != mint(0));
if (deg == -1) deg = (int)this->size();
FormalPowerSeries<mint> res(deg);
res[0] = {mint(1) / (*this)[0]};
for (int d = 1; d < deg; d <<= 1) {
FormalPowerSeries<mint> f(2 * d), g(2 * d);
for (int j = 0; j < min((int)this->size(), 2 * d); j++) f[j] = (*this)[j];
for (int j = 0; j < d; j++) g[j] = res[j];
f.ntt();
g.ntt();
for (int j = 0; j < 2 * d; j++) f[j] *= g[j];
f.intt();
for (int j = 0; j < d; j++) f[j] = 0;
f.ntt();
for (int j = 0; j < 2 * d; j++) f[j] *= g[j];
f.intt();
for (int j = d; j < min(2 * d, deg); j++) res[j] = -f[j];
}
return res.pre(deg);
}
template <typename mint>
FormalPowerSeries<mint> FormalPowerSeries<mint>::exp(int deg) const {
using fps = FormalPowerSeries<mint>;
assert((*this).size() == 0 || (*this)[0] == mint(0));
if (deg == -1) deg = this->size();
fps inv;
inv.reserve(deg + 1);
inv.push_back(mint(0));
inv.push_back(mint(1));
auto inplace_integral = [&](fps& F) -> void {
const int n = (int)F.size();
auto mod = mint::get_mod();
while ((int)inv.size() <= n) {
int i = inv.size();
inv.push_back((-inv[mod % i]) * (mod / i));
}
F.insert(begin(F), mint(0));
for (int i = 1; i <= n; i++) F[i] *= inv[i];
};
auto inplace_diff = [](fps& F) -> void {
if (F.empty()) return;
F.erase(begin(F));
mint coeff = 1, one = 1;
for (int i = 0; i < (int)F.size(); i++) {
F[i] *= coeff;
coeff += one;
}
};
fps b{1, 1 < (int)this->size() ? (*this)[1] : 0}, c{1}, z1, z2{1, 1};
for (int m = 2; m < deg; m *= 2) {
auto y = b;
y.resize(2 * m);
y.ntt();
z1 = z2;
fps z(m);
for (int i = 0; i < m; ++i) z[i] = y[i] * z1[i];
z.intt();
fill(begin(z), begin(z) + m / 2, mint(0));
z.ntt();
for (int i = 0; i < m; ++i) z[i] *= -z1[i];
z.intt();
c.insert(end(c), begin(z) + m / 2, end(z));
z2 = c;
z2.resize(2 * m);
z2.ntt();
fps x(begin(*this), begin(*this) + min<int>(this->size(), m));
x.resize(m);
inplace_diff(x);
x.push_back(mint(0));
x.ntt();
for (int i = 0; i < m; ++i) x[i] *= y[i];
x.intt();
x -= b.diff();
x.resize(2 * m);
for (int i = 0; i < m - 1; ++i) x[m + i] = x[i], x[i] = mint(0);
x.ntt();
for (int i = 0; i < 2 * m; ++i) x[i] *= z2[i];
x.intt();
x.pop_back();
inplace_integral(x);
for (int i = m; i < min<int>(this->size(), 2 * m); ++i) x[i] += (*this)[i];
fill(begin(x), begin(x) + m, mint(0));
x.ntt();
for (int i = 0; i < 2 * m; ++i) x[i] *= y[i];
x.intt();
b.insert(end(b), begin(x) + m, end(x));
}
return fps{begin(b), begin(b) + deg};
}
/**
* @brief NTT mod用FPSライブラリ
* @docs docs/fps/ntt-friendly-fps.md
*/
template <uint32_t mod>
struct LazyMontgomeryModInt {
using mint = LazyMontgomeryModInt;
using i32 = int32_t;
using u32 = uint32_t;
using u64 = uint64_t;
static constexpr u32 get_r() {
u32 ret = mod;
for (i32 i = 0; i < 4; ++i) ret *= 2 - mod * ret;
return ret;
}
static constexpr u32 r = get_r();
static constexpr u32 n2 = -u64(mod) % mod;
static_assert(mod < (1 << 30), "invalid, mod >= 2 ^ 30");
static_assert((mod & 1) == 1, "invalid, mod % 2 == 0");
static_assert(r * mod == 1, "this code has bugs.");
u32 a;
constexpr LazyMontgomeryModInt() : a(0) {}
constexpr LazyMontgomeryModInt(const int64_t &b)
: a(reduce(u64(b % mod + mod) * n2)){};
static constexpr u32 reduce(const u64 &b) {
return (b + u64(u32(b) * u32(-r)) * mod) >> 32;
}
constexpr mint &operator+=(const mint &b) {
if (i32(a += b.a - 2 * mod) < 0) a += 2 * mod;
return *this;
}
constexpr mint &operator-=(const mint &b) {
if (i32(a -= b.a) < 0) a += 2 * mod;
return *this;
}
constexpr mint &operator*=(const mint &b) {
a = reduce(u64(a) * b.a);
return *this;
}
constexpr mint &operator/=(const mint &b) {
*this *= b.inverse();
return *this;
}
constexpr mint operator+(const mint &b) const { return mint(*this) += b; }
constexpr mint operator-(const mint &b) const { return mint(*this) -= b; }
constexpr mint operator*(const mint &b) const { return mint(*this) *= b; }
constexpr mint operator/(const mint &b) const { return mint(*this) /= b; }
constexpr bool operator==(const mint &b) const {
return (a >= mod ? a - mod : a) == (b.a >= mod ? b.a - mod : b.a);
}
constexpr bool operator!=(const mint &b) const {
return (a >= mod ? a - mod : a) != (b.a >= mod ? b.a - mod : b.a);
}
constexpr mint operator-() const { return mint() - mint(*this); }
constexpr mint operator+() const { return mint(*this); }
constexpr mint pow(u64 n) const {
mint ret(1), mul(*this);
while (n > 0) {
if (n & 1) ret *= mul;
mul *= mul;
n >>= 1;
}
return ret;
}
constexpr mint inverse() const {
int x = get(), y = mod, u = 1, v = 0, t = 0, tmp = 0;
while (y > 0) {
t = x / y;
x -= t * y, u -= t * v;
tmp = x, x = y, y = tmp;
tmp = u, u = v, v = tmp;
}
return mint{u};
}
friend ostream &operator<<(ostream &os, const mint &b) {
return os << b.get();
}
friend istream &operator>>(istream &is, mint &b) {
int64_t t;
is >> t;
b = LazyMontgomeryModInt<mod>(t);
return (is);
}
constexpr u32 get() const {
u32 ret = reduce(a);
return ret >= mod ? ret - mod : ret;
}
static constexpr u32 get_mod() { return mod; }
};
using namespace std;
// コンストラクタの MAX に 「C(n, r) や fac(n) でクエリを投げる最大の n 」
// を入れると倍速くらいになる
// mod を超えて前計算して 0 割りを踏むバグは対策済み
template <typename T>
struct Binomial {
vector<T> f, g, h;
Binomial(int MAX = 0) {
assert(T::get_mod() != 0 && "Binomial<mint>()");
f.resize(1, T{1});
g.resize(1, T{1});
h.resize(1, T{1});
if (MAX > 0) extend(MAX + 1);
}
void extend(int m = -1) {
int n = f.size();
if (m == -1) m = n * 2;
m = min<int>(m, T::get_mod());
if (n >= m) return;
f.resize(m);
g.resize(m);
h.resize(m);
for (int i = n; i < m; i++) f[i] = f[i - 1] * T(i);
g[m - 1] = f[m - 1].inverse();
h[m - 1] = g[m - 1] * f[m - 2];
for (int i = m - 2; i >= n; i--) {
g[i] = g[i + 1] * T(i + 1);
h[i] = g[i] * f[i - 1];
}
}
T fac(int i) {
if (i < 0) return T(0);
while (i >= (int)f.size()) extend();
return f[i];
}
T finv(int i) {
if (i < 0) return T(0);
while (i >= (int)g.size()) extend();
return g[i];
}
T inv(int i) {
if (i < 0) return -inv(-i);
while (i >= (int)h.size()) extend();
return h[i];
}
T C(int n, int r) {
if (n < 0 || n < r || r < 0) return T(0);
return fac(n) * finv(n - r) * finv(r);
}
inline T operator()(int n, int r) { return C(n, r); }
template <typename I>
T multinomial(const vector<I>& r) {
static_assert(is_integral<I>::value == true);
int n = 0;
for (auto& x : r) {
if (x < 0) return T(0);
n += x;
}
T res = fac(n);
for (auto& x : r) res *= finv(x);
return res;
}
template <typename I>
T operator()(const vector<I>& r) {
return multinomial(r);
}
T C_naive(int n, int r) {
if (n < 0 || n < r || r < 0) return T(0);
T ret = T(1);
r = min(r, n - r);
for (int i = 1; i <= r; ++i) ret *= inv(i) * (n--);
return ret;
}
T P(int n, int r) {
if (n < 0 || n < r || r < 0) return T(0);
return fac(n) * finv(n - r);
}
// [x^r] 1 / (1-x)^n
T H(int n, int r) {
if (n < 0 || r < 0) return T(0);
return r == 0 ? 1 : C(n + r - 1, r);
}
};
// #include "fps/arbitrary-fps.hpp"
//
using namespace Nyaan;
using mint = LazyMontgomeryModInt<998244353>;
// using mint = LazyMontgomeryModInt<1000000007>;
using vm = vector<mint>;
using vvm = vector<vm>;
Binomial<mint> C;
using fps = FormalPowerSeries<mint>;
using namespace Nyaan;
using frac = fps_fraction<fps>;
using Mat = Matrix<fps>;
/*
void shrink(frac& prod) {
auto gc = PolyGCD(prod.p, prod.q);
gc *= gc[0].inverse();
prod.p /= gc;
prod.q /= gc;
prod.p.shrink();
prod.q.shrink();
}
// これを高速化
mint naive(int N, int M, int S, int T, vvi g) {
V<frac> dp(N);
auto dfs = [&](auto rc, int c, int p) -> frac {
frac s;
each(d, g[c]) {
if (d == p) continue;
rc(rc, d, c);
s += dp[d];
}
s = fps{1, -1} - s * fps{0, 0, 1};
swap(s.p, s.q);
return dp[c] = s;
};
dfs(dfs, T, -1);
rep(i, N) trc(dp[i]);
auto path = Path(g, S, T);
if (sz(path) - 1 > M) return 0;
frac prod = fps{1};
each(p, path) prod *= dp[p];
trc(prod);
// fps f = prod.p * (prod.q.inv(M + 10));
return LinearRecurrence(M - (sz(path) - 1), prod.q, prod.p);
}
*/
vvi g2;
frac light(int c);
V<frac> heavy(int c) {
if (g2[c].empty()) return V<frac>{frac{fps{0}, fps{1}}};
vector<frac> v, res = heavy(g2[c][0]);
v.push_back(frac{fps{0}, fps{1}});
rep1(i, sz(g2[c]) - 1) v.push_back(light(g2[c][i]));
while (v.size() >= 2u) {
vector<frac> w;
for (int i = 0; i + 1 < (int)v.size(); i += 2) w.push_back(v[i] + v[i + 1]);
if (v.size() & 1) w.push_back(v.back());
swap(v, w);
}
res.push_back(v.back());
return res;
}
frac light(int c) {
auto v = heavy(c);
Mat I(2);
I[0][0] = I[1][1] = fps{1};
Mat oya(2);
oya[0][1] = fps{1};
oya[1][0] = fps{0, 0, -1};
oya[1][1] = fps{1, -1};
V<Mat> ms{I, I};
each(f, v) {
Mat m(2);
m[0][0] = m[1][1] = f.q;
m[0][1] = f.p;
ms.push_back(m);
ms.push_back(oya);
}
reverse(all(ms));
while (sz(ms) >= 2) {
V<Mat> nx;
for (int i = 0; i + 1 < sz(ms); i += 2) nx.push_back(ms[i] * ms[i + 1]);
if (sz(ms) % 2) nx.push_back(ms.back());
ms = nx;
}
auto m = ms[0];
fps p = m[0][1];
fps q = m[1][1];
return frac{p, q};
}
mint calc(int N, int M, int S, int T, vvi g) {
auto path = Path(g, S, T);
g = rooted_tree(g, T);
HeavyLightDecomposition hld{g};
if (sz(path) - 1 > M) return 0;
{
set<int> pathset;
each(p, path) pathset.insert(p);
g2.resize(N);
rep(i, N) each(j, g[i]) {
if (pathset.count(j) == 0) g2[i].push_back(j);
}
}
// S, ..., T について問題を解く
V<frac> dp(sz(path));
rep(j, sz(path)) {
int c = path[j];
V<frac> v;
v.push_back(frac{fps{0}, fps{1}});
rep(i, sz(g2[c])) v.push_back(light(g2[c][i]));
while (v.size() >= 2u) {
vector<frac> w;
for (int i = 0; i + 1 < (int)v.size(); i += 2)
w.push_back(v[i] + v[i + 1]);
if (v.size() & 1) w.push_back(v.back());
swap(v, w);
}
dp[j] = v[0];
dp[j].p.shrink();
dp[j].q.shrink();
}
/*
// 一旦愚直を書く
frac prod{fps{1}, fps{1}};
{
frac s{fps{0}, fps{1}};
each(x, dp) {
s += x;
s = fps{1, -1} - s * fps{0, 0, 1};
swap(s.p, s.q);
prod *= s;
}
}
return LinearRecurrence(M - (sz(path) - 1), prod.q, prod.p);
*/
exit(1);
frac fs = fps{1, -1} - dp[0] * fps{0, 0, 1};
swap(fs.p, fs.q);
using Mat2 = Matrix<frac>;
V<Mat2> ms;
{
Mat2 m(2);
m[0][0] = m[1][1] = frac{fps{1}, fps{1}};
ms.push_back(m);
}
rep1(i, sz(path) - 1) {
Mat2 m(2);
m[0][1] = fps{1};
m[1][0] = fps{0, 0, -1};
m[1][1] = fps{1, -1} - dp[i] * fps{0, 0, 1};
ms.push_back(m);
}
reverse(all(ms));
while (sz(ms) >= 2) {
V<Mat2> nx;
for (int i = 0; i + 1 < sz(ms); i += 2) nx.push_back(ms[i] * ms[i + 1]);
if (sz(ms) % 2) nx.push_back(ms.back());
ms = nx;
}
auto m = ms[0];
frac ftp = fs.p * m[0][0] + fs.q * m[0][1];
frac ftq = fs.p * m[1][0] + fs.q * m[1][1];
trc(ftp, ftq);
frac f = fs.p / ftq;
trc(f);
return LinearRecurrence(M - (sz(path) - 1), f.q, f.p);
}
void q() {
ini(N, M, S, T);
auto g = graph(N);
S--, T--;
// out(naive(N, M, S, T, g));
out(calc(N, M, S, T, g));
}
void Nyaan::solve() {
int t = 1;
// in(t);
while (t--) q();
}