backup

src/bin/a.cc

@@ -1,5 +1,5 @@
 // #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
-#pragma GCC optimize("Ofast")
+#pragma GCC optimize("Ofast,unroll-loops")
 /************* This code requires C++17. ***************/
 
 #include<bits/stdc++.h>
@@ -279,9 +279,19 @@ return_t qpow(ll b, ll p) {
     else return half * half;
 }
 
-// Accurately find `i` 'th root of `n` (taking the floor)
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
 inline ll root(ll n, ll i) {
     ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
     while (l < r) {
@@ -527,24 +537,21 @@ void prep() {
 
 // __attribute__((target("popcnt")))
 void solve() {
-    read(int, n);
+    read(int, n, k);
-    vector<array<int, 2>> a(n);
+    readvec(int, a, n);
     for (int i = 0; i < n; ++i) {
-        cin >> a[i][0];
+        int has = 0;
-    }
-    for (int i = 0; i < n; ++i) {
-        cin >> a[i][1];
-    }
-    int res = -INF;
-    for (int i = 0; i < n; ++i) {
-        int curr = a[i][0] + a[i][1];
         for (int j = 0; j < n; ++j) {
-            if (j == i) continue;
+            if (i == j) continue;
-            curr += max(a[j][0], a[j][1]);
+            has |= abs(a[i] - a[j]) % k == 0;
         }
-        chmax(res, curr);
+        if (has == 0) {
+            cout << "YES\n";
+            cout << i + 1 << '\n';
+            return;
         }
-    cout << res << '\n';
+    }
+    cout << "NO\n";
 }
 
 int main() {

src/bin/arc-189a.cc

@@ -0,0 +1,611 @@
+// #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
+#pragma GCC optimize("Ofast,unroll-loops")
+/************* This code requires C++17. ***************/
+
+#include<bits/stdc++.h>
+using namespace std;
+
+/* macro helpers */
+#define __NARGS(...) std::tuple_size<decltype(std::make_tuple(__VA_ARGS__))>::value
+#define __DECOMPOSE_S(a, x) auto x = a;
+#define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
+constexpr void __() {}
+#define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
+#define __as_typeof(container) remove_reference<decltype(container)>::type
+template <typename T> struct argument_type;
+template <typename T, typename U> struct argument_type<T(U)> { using type = U; };
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937_64 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define adj(ch, n) __AS_PROCEDURE(vector<vector<int>> ch((n) + 1);)
+#define edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v), ch[v].push_back(u);)
+#define edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__), ch[v].emplace_back(u, __VA_ARGS__);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__);)
+template <typename T, typename Iterator> pair<size_t, map<T, size_t>> discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    map<T, size_t> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, safe_hash>> unordered_discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    unordered_map<T, size_t, safe_hash> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+
+/* io */
+#define untie __AS_PROCEDURE(ios_base::sync_with_stdio(0), cin.tie(NULL))
+
+// add declarations to avoid cyclic dependency
+template<typename T, typename U> istream& operator>>(istream&, pair<T, U>&);
+template<typename T, typename U> ostream& operator<<(ostream&, const pair<T, U>&);
+template<typename T, size_t N> istream& operator>>(istream&, array<T, N>&);
+template <typename T, size_t N> ostream& operator<<(ostream&, const array<T, N>&);
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>&, const std::tuple<Args...>&);
+template<typename T> ostream& operator<<(ostream&, const vector<T>&);
+std::ostream& operator<<(std::ostream&, const int128&);
+
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename T, size_t N> istream& operator>>(istream& in, array<T, N>& a) {
+    for (size_t i = 0; i < N; ++i) in >> a[i];
+    return in;
+}
+template <typename T, size_t N> ostream& operator<<(ostream& out, const array<T, N>& a) {
+    for (auto&& i : a) out << i << ' ';
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+template<typename T> void __read(T& x) { cin >> x; }
+template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __read(args...); }
+#define read(t, ...) __AS_PROCEDURE(argument_type<void(t)>::type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a(n); for (auto& x : a) cin >> x;)
+#define readvec1(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a((n) + 1); copy_n(ii<argument_type<void(t)>::type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+
+/* pops */
+template <typename Container>
+inline auto poptop(Container& q) {
+    auto ret = q.top();
+    q.pop();
+    return ret;
+}
+template <typename Container>
+inline auto popback(Container& q) {
+    auto ret = q.back();
+    q.pop_back();
+    return ret;
+}
+template <typename Container>
+inline auto popfront(Container& q) {
+    auto ret = q.front();
+    q.pop_front();
+    return ret;
+}
+
+/* math */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
+inline ll root(ll n, ll i) {
+    ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
+    while (l < r) {
+        ll mid = l + r + 1 >> 1;
+        if (qpow<int128>(mid, i) <= n) {
+            l = mid;
+        } else {
+            r = mid - 1;
+        }
+    }
+    return l;
+}
+#pragma GCC diagnostic pop
+
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+#define fastcomb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] * factrev[k] * factrev[(n) - (k)])
+
+__attribute__((target("lzcnt")))
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+template <typename T>
+T mygcd(T a, T b) { return b == 0 ? a : mygcd(b, a % b); }
+
+void __exgcd(ll a, ll b, ll& x, ll& y) {
+  if (b == 0) {
+    x = 1, y = 0;
+    return;
+  }
+  __exgcd(b, a % b, y, x);
+  y -= a / b * x;
+}
+
+ll inverse(ll a, ll b) {
+    ll x, y;
+    __exgcd(a, b, x, y);
+    return mod(x, b);
+}
+
+vector<tuple<int, int, ll>> decompose(ll x) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+    int n = (int)t.length();
+    vector<int> pi(n);
+    for (int i = 1; i < n; i++) {
+        int j = pi[i - 1];
+        while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+        if (t[i] == t[j]) j++;
+        pi[i] = j;
+    }
+    return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+    string cur = t + '#' + s;
+    int sz1 = s.size(), sz2 = t.size();
+    vector<int> v;
+    vector<int> lps = calc_next(cur);
+    for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+        if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+    }
+    return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    MLL& operator+=(const MLL& rhs) { return *this = *this + rhs; }
+    MLL& operator-=(const MLL& rhs) { return *this = *this - rhs; }
+    MLL& operator*=(const MLL& rhs) { return *this = *this * rhs; }
+    MLL& operator/=(const MLL& rhs) { return *this = *this / rhs; }
+    MLL& operator%=(const MLL& rhs) { return *this = *this % rhs; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) ([&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);})
+#define expr(ret, ...) ([&] (__VA_ARGS__) { return (ret); })
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+namespace detail {
+    template <typename T, std::size_t...Is>
+    constexpr std::array<T, sizeof...(Is)>
+    make_array(const T& value, std::index_sequence<Is...>) {
+        return {{(static_cast<void>(Is), value)...}};
+    }
+}
+
+template <typename T, std::size_t N>
+constexpr std::array<T, N> __initarray(const T& value) {
+    return detail::make_array(value, std::make_index_sequence<N>());
+}
+/*******************************************************/
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+using mll = MLL<PRIME>;
+constexpr int N = 2e5 + 10;
+mll fact[N];
+
+void prep() {
+    fact[0] = 1;
+    for (int i = 1; i < N; ++i) {
+        fact[i] = fact[i - 1] * i;
+    }
+}
+
+// __attribute__((target("popcnt")))
+void solve() {
+    read(int, n);
+    vector<mll> df(n + 1);
+    df[0] = df[1] = 1;
+    for (int i = 2; i <= n; ++i) {
+        df[i] = df[i - 2] * i;
+    }
+    readvec(int, a, n);
+    int s = 0;
+    vector<int> segs;
+    int rem = 0;
+    for (int i = 1; i < n; ++i) {
+        if (a[i] != a[s]) {
+            if ((i - s) % 2 == 0 or s == 0 and a[s] == 0) {
+                cout << "0\n";
+                return;
+            }
+            if (i - s != 1) {
+                segs.emplace_back(i - s);
+                rem += (i - s) / 2;
+            }
+            s = i;
+        }
+    }
+    if ((n - s) % 2 == 0 or s == 0 and a[s] == 0) {
+        cout << "0\n";
+        return;
+    }
+    if (n - s != 1) {
+        segs.emplace_back(n - s);
+        rem += (n - s) / 2;
+    }
+    mll res = 1;
+    for (auto&& x : segs) {
+        res *= df[x - 2] * comb(rem, x / 2);
+        rem -= x / 2;
+    }
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/b.cc

@@ -1,5 +1,5 @@
 // #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
-#pragma GCC optimize("Ofast")
+#pragma GCC optimize("Ofast,unroll-loops")
 /************* This code requires C++17. ***************/
 
 #include<bits/stdc++.h>
@@ -279,9 +279,19 @@ return_t qpow(ll b, ll p) {
     else return half * half;
 }
 
-// Accurately find `i` 'th root of `n` (taking the floor)
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
 inline ll root(ll n, ll i) {
     ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
     while (l < r) {
@@ -528,39 +538,13 @@ void prep() {
 // __attribute__((target("popcnt")))
 void solve() {
     read(int, n);
-    readvec(int, a, n);
+    int res = 1;
-    vector<int> stack;
+    int curr = 1;
-    vector<bool> mark(n);
+    while (curr < n) {
-    for (int i = 0; i < n; ++i) {
+        res += 1;
-        while (stack.size() and a[stack.back()] > a[i]) {
+        curr += curr + 2;
-            mark[stack.back()] = 0;
-            stack.pop_back();
     }
-        mark[i] = 1;
+    cout << res << '\n';
-        stack.emplace_back(i);
-    }
-    int mn = INF;
-    for (int i = 0; i < n; ++i) {
-        if (not mark[i]) {
-            chmin(mn, a[i] + 1);
-        }
-    }
-    while (stack.size() and a[stack.back()] > mn) {
-        mark[stack.back()] = 0;
-        stack.pop_back();
-    }
-    multiset<int> rem;
-    for (int i = 0; i < n; ++i) {
-        if (mark[i]) {
-            cout << a[i] << ' ';
-        } else {
-            rem.emplace(a[i] + 1);
-        }
-    }
-    for (auto&& x : rem) {
-        cout << x << ' ';
-    }
-    cout << '\n';
 }
 
 int main() {

src/bin/c.cc

@@ -1,5 +1,5 @@
 // #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
-#pragma GCC optimize("Ofast")
+#pragma GCC optimize("Ofast,unroll-loops")
 /************* This code requires C++17. ***************/
 
 #include<bits/stdc++.h>
@@ -279,9 +279,19 @@ return_t qpow(ll b, ll p) {
     else return half * half;
 }
 
-// Accurately find `i` 'th root of `n` (taking the floor)
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
 inline ll root(ll n, ll i) {
     ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
     while (l < r) {
@@ -525,188 +535,37 @@ void dump_ignore() {}
 void prep() {
 }
 
-template<typename Addable_Info_t, typename Tag_t, typename Sequence = std::vector<Addable_Info_t>> class segtree {
-private:
-    using size_type = uint64_t;
-    using info_type = Addable_Info_t;
-    using tag_type = Tag_t;
-    size_type _max;
-    vector<info_type> d;
-    vector<tag_type> b;
-    void pull(size_type p) {
-        d[p] = d[p * 2] + d[p * 2 + 1];
-    }
-    void push(size_type p, size_type left_len, size_type right_len) {
-        d[p * 2].apply(b[p], left_len), d[p * 2 + 1].apply(b[p], right_len);
-        b[p * 2].apply(b[p]), b[p * 2 + 1].apply(b[p]);
-        b[p] = tag_type();
-    }
-    void set(size_type s, size_type t, size_type p, size_type x, const info_type& c) {
-        if (s == t) {
-            d[p] = c;
-            return;
-        }
-        size_type m = s + (t - s >> 1);
-        if (s != t) push(p, m - s + 1, t - m);
-        if (x <= m) set(s, m, p * 2, x, c);
-        else set(m + 1, t, p * 2 + 1, x, c);
-        pull(p);
-    }
-
-    void range_apply(size_type s, size_type t, size_type p, size_type l, size_type r, const tag_type& c) {
-        if (l <= s && t <= r) {
-            d[p].apply(c, t - s + 1);
-            b[p].apply(c);
-            return;
-        }
-        size_type m = s + (t - s >> 1);
-        push(p, m - s + 1, t - m);
-        if (l <= m) range_apply(s, m, p * 2, l, r, c);
-        if (r > m)  range_apply(m + 1, t, p * 2 + 1, l, r, c);
-        pull(p);
-    }
-    info_type range_query(size_type s, size_type t, size_type p, size_type l, size_type r) {
-        if (l <= s && t <= r) {
-            return d[p];
-        }
-        size_type m = s + (t - s >> 1);
-        info_type res = {};
-        push(p, m - s + 1, t - m);
-        if (l <= m) res = res + range_query(s, m, p * 2, l, r);
-        if (r > m)  res = res + range_query(m + 1, t, p * 2 + 1, l, r);
-        return res;
-    }
-    int binary_search(size_type s, size_type t, size_type p, int val) {
-        if (s == t) {
-            return d[p].val >= val ? s : INF;
-        }
-        size_type m = s + (t - s >> 1);
-        int right = d[p * 2 + 1].val;
-        if (right >= val) return binary_search(m + 1, t, p * 2 + 1, val);
-        else return binary_search(s, m, p * 2, val - right);
-    }
-    void build(const Sequence& a, size_type s, size_type t, size_type p) {
-        if (s == t) {
-            d[p] = a[s];
-            return;
-        }
-        int m = s + (t - s >> 1);
-        build(a, s, m, p * 2);
-        build(a, m + 1, t, p * 2 + 1);
-        pull(p);
-    }
-public:
-    segtree(size_type __max) : d(4 * __max), b(4 * __max), _max(__max - 1) {}
-    segtree(const Sequence& a) : segtree(a.size()) {
-        build(a, {}, _max, 1);
-    }
-    void set(size_type i, const info_type& c) {
-        set({}, _max, 1, i, c);
-    }
-
-    void range_apply(size_type l, size_type r, const tag_type& c) {
-        range_apply({}, _max, 1, l, r, c);
-    }
-    void apply(size_type i, const tag_type& c) {
-        range_apply(i, i, c);
-    }
-    info_type range_query(size_type l, size_type r) {
-        return range_query({}, _max, 1, l, r);
-    }
-    int binary_search(int val) {
-        return binary_search({}, _max, 1, val);
-    }
-    info_type query(size_type i) {
-        return range_query(i, i);
-    }
-    Sequence serialize() {
-        Sequence res = {};
-        for (size_type i = 0; i <= _max; ++i) {
-            res.push_back(query(i));
-        }
-        return res;
-    }
-    const vector<info_type>& get_d() {
-        return d;
-    }
-};
-struct Tag {
-    int val = 0;
-    void apply(const Tag& rhs) {
-        val += rhs.val;
-    }
-};
-struct Info {
-    int val = 0;
-    void apply(const Tag& rhs, size_t len) {
-        val += rhs.val * len;
-    }
-};
-Info operator+(const Info &a, const Info &b) {
-    return {a.val + b.val};
-}
-
 // __attribute__((target("popcnt")))
 void solve() {
     read(int, n);
-    vector<array<int, 2>> a(n);
+    read(ll, k);
-    vector<int> r, c;
+    k -= 1;
-    for (int i = 0; i < n; ++i) {
+    vector<int> res(n);
-        read(int, x, y);
+    int l = 0, r = n - 1;
-        a[i] = { x, y };
+    for (int i = n - 2; ~i; --i) {
-        r.emplace_back(x);
+        int bit;
-        c.emplace_back(y);
+        if (i > 63) {
-    }
+            bit = 0;
-    sort(r.begin(), r.end());
-    int mr = unique(r.begin(), r.end()) - r.begin();
-    r.resize(mr);
-    auto gr = [&] (int x) { return lower_bound(r.begin(), r.end(), x) - r.begin(); };
-    sort(c.begin(), c.end());
-    int mc = unique(c.begin(), c.end()) - c.begin();
-    c.resize(mc);
-    auto gc = [&] (int x) { return lower_bound(c.begin(), c.end(), x) - c.begin(); };
-    vector<vector<int>> open(mr);
-    for (int i = 0; i < n; ++i) {
-        open[gr(a[i][0])].emplace_back(i);
-    }
-    auto check = [&] (int cnt) -> optional<pii> {
-        if (cnt == 0) return { { 0, 0 } };
-        segtree<Info, Tag> up(mc), down(mc);
-        int tu = 0, td = 0;
-        for (int i = 0; i < n; ++i) {
-            up.apply(gc(a[i][1]), { 1 });
-            tu += 1;
-        }
-        for (int i = mr - 1; ~i; --i) {
-            for (auto&& idx : open[i]) {
-                up.apply(gc(a[idx][1]), { -1 });
-                down.apply(gc(a[idx][1]), { 1 });
-                tu -= 1;
-                td += 1;
-            }
-            int p = up.binary_search(cnt);
-            int q = down.binary_search(cnt);
-            if (p < INF and q < INF) {
-                p = min(p, q);
-                if (min(tu - up.range_query(p, mc - 1).val, td - down.range_query(p, mc - 1).val) >= cnt) {
-                    return { { r[i], c[p] } };
-                }
-            }
-        }
-        return nullopt;
-    };
-    int left = 0, right = n / 4;
-    while (left < right) {
-        int mid = left + right + 1 >> 1;
-        if (check(mid)) {
-            left = mid;
         } else {
-            right = mid - 1;
+            bit = k >> i & 1;
+            k ^= ll(bit) << i;
+        }
+        if (bit == 0) {
+            res[l++] = n - 1 - i;
+        } else {
+            res[r--] = n - 1 - i;
         }
     }
-    auto [x, y] = *check(left);
+    if (k) {
-    cout << left << '\n';
+        cout << -1 << '\n';
-    cout << x << ' ' << y << '\n';
+    } else {
+        assert(l == r);
+        res[l] = n;
+        for (auto&& x : res) {
+            cout << x << ' ';
+        }
+        cout << '\n';
+    }
 }
 
 int main() {

src/bin/cf-2040d.cc

@@ -0,0 +1,623 @@
+// #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
+#pragma GCC optimize("Ofast,unroll-loops")
+/************* This code requires C++17. ***************/
+
+#include<bits/stdc++.h>
+using namespace std;
+
+/* macro helpers */
+#define __NARGS(...) std::tuple_size<decltype(std::make_tuple(__VA_ARGS__))>::value
+#define __DECOMPOSE_S(a, x) auto x = a;
+#define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
+constexpr void __() {}
+#define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
+#define __as_typeof(container) remove_reference<decltype(container)>::type
+template <typename T> struct argument_type;
+template <typename T, typename U> struct argument_type<T(U)> { using type = U; };
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937_64 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define adj(ch, n) __AS_PROCEDURE(vector<vector<int>> ch((n) + 1);)
+#define edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v), ch[v].push_back(u);)
+#define edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__), ch[v].emplace_back(u, __VA_ARGS__);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__);)
+template <typename T, typename Iterator> pair<size_t, map<T, size_t>> discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    map<T, size_t> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, safe_hash>> unordered_discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    unordered_map<T, size_t, safe_hash> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+
+/* io */
+#define untie __AS_PROCEDURE(ios_base::sync_with_stdio(0), cin.tie(NULL))
+
+// add declarations to avoid cyclic dependency
+template<typename T, typename U> istream& operator>>(istream&, pair<T, U>&);
+template<typename T, typename U> ostream& operator<<(ostream&, const pair<T, U>&);
+template<typename T, size_t N> istream& operator>>(istream&, array<T, N>&);
+template <typename T, size_t N> ostream& operator<<(ostream&, const array<T, N>&);
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>&, const std::tuple<Args...>&);
+template<typename T> ostream& operator<<(ostream&, const vector<T>&);
+std::ostream& operator<<(std::ostream&, const int128&);
+
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename T, size_t N> istream& operator>>(istream& in, array<T, N>& a) {
+    for (size_t i = 0; i < N; ++i) in >> a[i];
+    return in;
+}
+template <typename T, size_t N> ostream& operator<<(ostream& out, const array<T, N>& a) {
+    for (auto&& i : a) out << i << ' ';
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+template<typename T> void __read(T& x) { cin >> x; }
+template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __read(args...); }
+#define read(t, ...) __AS_PROCEDURE(argument_type<void(t)>::type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a(n); for (auto& x : a) cin >> x;)
+#define readvec1(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a((n) + 1); copy_n(ii<argument_type<void(t)>::type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+
+/* pops */
+template <typename Container>
+inline auto poptop(Container& q) {
+    auto ret = q.top();
+    q.pop();
+    return ret;
+}
+template <typename Container>
+inline auto popback(Container& q) {
+    auto ret = q.back();
+    q.pop_back();
+    return ret;
+}
+template <typename Container>
+inline auto popfront(Container& q) {
+    auto ret = q.front();
+    q.pop_front();
+    return ret;
+}
+
+/* math */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
+inline ll root(ll n, ll i) {
+    ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
+    while (l < r) {
+        ll mid = l + r + 1 >> 1;
+        if (qpow<int128>(mid, i) <= n) {
+            l = mid;
+        } else {
+            r = mid - 1;
+        }
+    }
+    return l;
+}
+#pragma GCC diagnostic pop
+
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+#define fastcomb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] * factrev[k] * factrev[(n) - (k)])
+
+__attribute__((target("lzcnt")))
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+template <typename T>
+T mygcd(T a, T b) { return b == 0 ? a : mygcd(b, a % b); }
+
+void __exgcd(ll a, ll b, ll& x, ll& y) {
+  if (b == 0) {
+    x = 1, y = 0;
+    return;
+  }
+  __exgcd(b, a % b, y, x);
+  y -= a / b * x;
+}
+
+ll inverse(ll a, ll b) {
+    ll x, y;
+    __exgcd(a, b, x, y);
+    return mod(x, b);
+}
+
+vector<tuple<int, int, ll>> decompose(ll x) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+    int n = (int)t.length();
+    vector<int> pi(n);
+    for (int i = 1; i < n; i++) {
+        int j = pi[i - 1];
+        while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+        if (t[i] == t[j]) j++;
+        pi[i] = j;
+    }
+    return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+    string cur = t + '#' + s;
+    int sz1 = s.size(), sz2 = t.size();
+    vector<int> v;
+    vector<int> lps = calc_next(cur);
+    for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+        if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+    }
+    return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    MLL& operator+=(const MLL& rhs) { return *this = *this + rhs; }
+    MLL& operator-=(const MLL& rhs) { return *this = *this - rhs; }
+    MLL& operator*=(const MLL& rhs) { return *this = *this * rhs; }
+    MLL& operator/=(const MLL& rhs) { return *this = *this / rhs; }
+    MLL& operator%=(const MLL& rhs) { return *this = *this % rhs; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) ([&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);})
+#define expr(ret, ...) ([&] (__VA_ARGS__) { return (ret); })
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+namespace detail {
+    template <typename T, std::size_t...Is>
+    constexpr std::array<T, sizeof...(Is)>
+    make_array(const T& value, std::index_sequence<Is...>) {
+        return {{(static_cast<void>(Is), value)...}};
+    }
+}
+
+template <typename T, std::size_t N>
+constexpr std::array<T, N> __initarray(const T& value) {
+    return detail::make_array(value, std::make_index_sequence<N>());
+}
+/*******************************************************/
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+constexpr int N = 4e5 + 10;
+bool not_prime[N + 1];
+
+void soe(int n) {
+    vector<int> res;
+    for (int i = 2; i <= n; ++i) {
+        if (not not_prime[i]) {
+            res.emplace_back(i);
+        }
+        for (auto&& x : res) {
+            if (i * x > n) break;
+            not_prime[i * x] = 1;
+            if (i % x == 0) break;
+        }
+    }
+}
+
+void prep() {
+    soe(N);
+}
+
+// __attribute__((target("popcnt")))
+void solve() {
+    read(int, n);
+    adj(ch, n);
+    for (int i = 0; i < n - 1; ++i) {
+        read(int, u, v);
+        edge(ch, u, v);
+    }
+    array<vector<int>, 2> bk;
+    vector<int> fa(n + 1);
+    {
+        auto dfs = [&] (auto dfs, int v, int pa, int par) -> void {
+            fa[v] = pa;
+            bk[par].emplace_back(v);
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                dfs(dfs, u, v, 1 - par);
+            }
+        };
+        dfs(dfs, 1, 0, 0);
+    }
+    vector<int> res(n + 1);
+    int curr = 2;
+    for (auto&& v : bk[0]) {
+        res[v] = curr;
+        curr += 2;
+    }
+    curr = 2 * n;
+    for (auto&& v : bk[1]) {
+        res[v] = curr;
+        curr -= 2;
+    }
+    for (int i = 2; i <= n; ++i) {
+        if (abs(res[fa[i]] - res[i]) == 2) {
+            res[i] = res[fa[i]] > res[i] ? res[i] + 1 : res[i] - 1;
+            break;
+        }
+    }
+    for (int i = 1; i <= n; ++i) {
+        cout << res[i] << " \n"[i == n];
+    }
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-2040e.cc

@@ -0,0 +1,608 @@
+// #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
+#pragma GCC optimize("Ofast,unroll-loops")
+/************* This code requires C++17. ***************/
+
+#include<bits/stdc++.h>
+using namespace std;
+
+/* macro helpers */
+#define __NARGS(...) std::tuple_size<decltype(std::make_tuple(__VA_ARGS__))>::value
+#define __DECOMPOSE_S(a, x) auto x = a;
+#define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
+constexpr void __() {}
+#define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
+#define __as_typeof(container) remove_reference<decltype(container)>::type
+template <typename T> struct argument_type;
+template <typename T, typename U> struct argument_type<T(U)> { using type = U; };
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937_64 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define adj(ch, n) __AS_PROCEDURE(vector<vector<int>> ch((n) + 1);)
+#define edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v), ch[v].push_back(u);)
+#define edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__), ch[v].emplace_back(u, __VA_ARGS__);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__);)
+template <typename T, typename Iterator> pair<size_t, map<T, size_t>> discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    map<T, size_t> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, safe_hash>> unordered_discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    unordered_map<T, size_t, safe_hash> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+
+/* io */
+#define untie __AS_PROCEDURE(ios_base::sync_with_stdio(0), cin.tie(NULL))
+
+// add declarations to avoid cyclic dependency
+template<typename T, typename U> istream& operator>>(istream&, pair<T, U>&);
+template<typename T, typename U> ostream& operator<<(ostream&, const pair<T, U>&);
+template<typename T, size_t N> istream& operator>>(istream&, array<T, N>&);
+template <typename T, size_t N> ostream& operator<<(ostream&, const array<T, N>&);
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>&, const std::tuple<Args...>&);
+template<typename T> ostream& operator<<(ostream&, const vector<T>&);
+std::ostream& operator<<(std::ostream&, const int128&);
+
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename T, size_t N> istream& operator>>(istream& in, array<T, N>& a) {
+    for (size_t i = 0; i < N; ++i) in >> a[i];
+    return in;
+}
+template <typename T, size_t N> ostream& operator<<(ostream& out, const array<T, N>& a) {
+    for (auto&& i : a) out << i << ' ';
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+template<typename T> void __read(T& x) { cin >> x; }
+template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __read(args...); }
+#define read(t, ...) __AS_PROCEDURE(argument_type<void(t)>::type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a(n); for (auto& x : a) cin >> x;)
+#define readvec1(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a((n) + 1); copy_n(ii<argument_type<void(t)>::type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+
+/* pops */
+template <typename Container>
+inline auto poptop(Container& q) {
+    auto ret = q.top();
+    q.pop();
+    return ret;
+}
+template <typename Container>
+inline auto popback(Container& q) {
+    auto ret = q.back();
+    q.pop_back();
+    return ret;
+}
+template <typename Container>
+inline auto popfront(Container& q) {
+    auto ret = q.front();
+    q.pop_front();
+    return ret;
+}
+
+/* math */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
+inline ll root(ll n, ll i) {
+    ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
+    while (l < r) {
+        ll mid = l + r + 1 >> 1;
+        if (qpow<int128>(mid, i) <= n) {
+            l = mid;
+        } else {
+            r = mid - 1;
+        }
+    }
+    return l;
+}
+#pragma GCC diagnostic pop
+
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+#define fastcomb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] * factrev[k] * factrev[(n) - (k)])
+
+__attribute__((target("lzcnt")))
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+template <typename T>
+T mygcd(T a, T b) { return b == 0 ? a : mygcd(b, a % b); }
+
+void __exgcd(ll a, ll b, ll& x, ll& y) {
+  if (b == 0) {
+    x = 1, y = 0;
+    return;
+  }
+  __exgcd(b, a % b, y, x);
+  y -= a / b * x;
+}
+
+ll inverse(ll a, ll b) {
+    ll x, y;
+    __exgcd(a, b, x, y);
+    return mod(x, b);
+}
+
+vector<tuple<int, int, ll>> decompose(ll x) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+    int n = (int)t.length();
+    vector<int> pi(n);
+    for (int i = 1; i < n; i++) {
+        int j = pi[i - 1];
+        while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+        if (t[i] == t[j]) j++;
+        pi[i] = j;
+    }
+    return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+    string cur = t + '#' + s;
+    int sz1 = s.size(), sz2 = t.size();
+    vector<int> v;
+    vector<int> lps = calc_next(cur);
+    for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+        if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+    }
+    return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    MLL& operator+=(const MLL& rhs) { return *this = *this + rhs; }
+    MLL& operator-=(const MLL& rhs) { return *this = *this - rhs; }
+    MLL& operator*=(const MLL& rhs) { return *this = *this * rhs; }
+    MLL& operator/=(const MLL& rhs) { return *this = *this / rhs; }
+    MLL& operator%=(const MLL& rhs) { return *this = *this % rhs; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) ([&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);})
+#define expr(ret, ...) ([&] (__VA_ARGS__) { return (ret); })
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+namespace detail {
+    template <typename T, std::size_t...Is>
+    constexpr std::array<T, sizeof...(Is)>
+    make_array(const T& value, std::index_sequence<Is...>) {
+        return {{(static_cast<void>(Is), value)...}};
+    }
+}
+
+template <typename T, std::size_t N>
+constexpr std::array<T, N> __initarray(const T& value) {
+    return detail::make_array(value, std::make_index_sequence<N>());
+}
+/*******************************************************/
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+// __attribute__((target("popcnt")))
+void solve() {
+    read(int, n, q);
+    adj(ch, n);
+    for (int i = 0; i < n - 1; ++i) {
+        read(int, u, v);
+        edge(ch, u, v);
+    }
+    vector<int> fa(n + 1);
+    vector dp(n + 1, vector<ll>(n + 1, INFLL));
+    dp[1].assign(n + 1, 0);
+    {
+        auto dfs = [&] (auto dfs, int v, int pa) -> void {
+            fa[v] = pa;
+            if (v != 1) {
+                for (int j = 0; j <= n; ++j) {
+                    if (fa[v] != 1) {
+                        // use a coin
+                        if (j > 0) {
+                            chmin(dp[v][j], dp[fa[fa[v]]][j - 1] + 2);
+                        }
+                        // don't use a coin
+                        chmin(dp[v][j], (dp[fa[fa[v]]][j] + 2) + 2 * (ch[v].size() - 1));
+                    } else {
+                        // use a coin
+                        if (j > 0) {
+                            chmin(dp[v][j], 1);
+                        }
+                        // don't use a coin
+                        chmin(dp[v][j], (dp[fa[v]][j] + 1) + 2 * (ch[v].size() - 1));
+                    }
+                }
+            }
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                dfs(dfs, u, v);
+            }
+        };
+        dfs(dfs, 1, 0);
+    }
+    while (q--) {
+        read(int, v, p);
+        cout << dp[fa[v]][p] + 1 << '\n';
+    }
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-908f.cc

@@ -0,0 +1,652 @@
+// #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
+#pragma GCC optimize("Ofast,unroll-loops")
+/************* This code requires C++17. ***************/
+
+#include<bits/stdc++.h>
+using namespace std;
+
+/* macro helpers */
+#define __NARGS(...) std::tuple_size<decltype(std::make_tuple(__VA_ARGS__))>::value
+#define __DECOMPOSE_S(a, x) auto x = a;
+#define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
+constexpr void __() {}
+#define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
+#define __as_typeof(container) remove_reference<decltype(container)>::type
+template <typename T> struct argument_type;
+template <typename T, typename U> struct argument_type<T(U)> { using type = U; };
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937_64 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define adj(ch, n) __AS_PROCEDURE(vector<vector<int>> ch((n) + 1);)
+#define edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v), ch[v].push_back(u);)
+#define edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__), ch[v].emplace_back(u, __VA_ARGS__);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__);)
+template <typename T, typename Iterator> pair<size_t, map<T, size_t>> discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    map<T, size_t> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, safe_hash>> unordered_discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    unordered_map<T, size_t, safe_hash> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+
+/* io */
+#define untie __AS_PROCEDURE(ios_base::sync_with_stdio(0), cin.tie(NULL))
+
+// add declarations to avoid cyclic dependency
+template<typename T, typename U> istream& operator>>(istream&, pair<T, U>&);
+template<typename T, typename U> ostream& operator<<(ostream&, const pair<T, U>&);
+template<typename T, size_t N> istream& operator>>(istream&, array<T, N>&);
+template <typename T, size_t N> ostream& operator<<(ostream&, const array<T, N>&);
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>&, const std::tuple<Args...>&);
+template<typename T> ostream& operator<<(ostream&, const vector<T>&);
+std::ostream& operator<<(std::ostream&, const int128&);
+
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename T, size_t N> istream& operator>>(istream& in, array<T, N>& a) {
+    for (size_t i = 0; i < N; ++i) in >> a[i];
+    return in;
+}
+template <typename T, size_t N> ostream& operator<<(ostream& out, const array<T, N>& a) {
+    for (auto&& i : a) out << i << ' ';
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+template<typename T> void __read(T& x) { cin >> x; }
+template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __read(args...); }
+#define read(t, ...) __AS_PROCEDURE(argument_type<void(t)>::type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a(n); for (auto& x : a) cin >> x;)
+#define readvec1(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a((n) + 1); copy_n(ii<argument_type<void(t)>::type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+
+/* pops */
+template <typename Container>
+inline auto poptop(Container& q) {
+    auto ret = q.top();
+    q.pop();
+    return ret;
+}
+template <typename Container>
+inline auto popback(Container& q) {
+    auto ret = q.back();
+    q.pop_back();
+    return ret;
+}
+template <typename Container>
+inline auto popfront(Container& q) {
+    auto ret = q.front();
+    q.pop_front();
+    return ret;
+}
+
+/* math */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
+inline ll root(ll n, ll i) {
+    ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
+    while (l < r) {
+        ll mid = l + r + 1 >> 1;
+        if (qpow<int128>(mid, i) <= n) {
+            l = mid;
+        } else {
+            r = mid - 1;
+        }
+    }
+    return l;
+}
+#pragma GCC diagnostic pop
+
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+#define fastcomb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] * factrev[k] * factrev[(n) - (k)])
+
+__attribute__((target("lzcnt")))
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+template <typename T>
+T mygcd(T a, T b) { return b == 0 ? a : mygcd(b, a % b); }
+
+void __exgcd(ll a, ll b, ll& x, ll& y) {
+  if (b == 0) {
+    x = 1, y = 0;
+    return;
+  }
+  __exgcd(b, a % b, y, x);
+  y -= a / b * x;
+}
+
+ll inverse(ll a, ll b) {
+    ll x, y;
+    __exgcd(a, b, x, y);
+    return mod(x, b);
+}
+
+vector<tuple<int, int, ll>> decompose(ll x) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+    int n = (int)t.length();
+    vector<int> pi(n);
+    for (int i = 1; i < n; i++) {
+        int j = pi[i - 1];
+        while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+        if (t[i] == t[j]) j++;
+        pi[i] = j;
+    }
+    return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+    string cur = t + '#' + s;
+    int sz1 = s.size(), sz2 = t.size();
+    vector<int> v;
+    vector<int> lps = calc_next(cur);
+    for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+        if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+    }
+    return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    MLL& operator+=(const MLL& rhs) { return *this = *this + rhs; }
+    MLL& operator-=(const MLL& rhs) { return *this = *this - rhs; }
+    MLL& operator*=(const MLL& rhs) { return *this = *this * rhs; }
+    MLL& operator/=(const MLL& rhs) { return *this = *this / rhs; }
+    MLL& operator%=(const MLL& rhs) { return *this = *this % rhs; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) ([&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);})
+#define expr(ret, ...) ([&] (__VA_ARGS__) { return (ret); })
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+namespace detail {
+    template <typename T, std::size_t...Is>
+    constexpr std::array<T, sizeof...(Is)>
+    make_array(const T& value, std::index_sequence<Is...>) {
+        return {{(static_cast<void>(Is), value)...}};
+    }
+}
+
+template <typename T, std::size_t N>
+constexpr std::array<T, N> __initarray(const T& value) {
+    return detail::make_array(value, std::make_index_sequence<N>());
+}
+/*******************************************************/
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+// __attribute__((target("popcnt")))
+void solve() {
+    read(int, n);
+    readvec((pair<int, char>), a, n);
+    vector<int> pre(n);
+    pre[0] = a[0].second == 'G' ? 0 : -1;
+    for (int i = 1; i < n; ++i) {
+        if (a[i].second == 'G') {
+            pre[i] = i;
+        } else {
+            pre[i] = pre[i - 1];
+        }
+    }
+    vector<int> nxt(n);
+    nxt[n - 1] = a[n - 1].second == 'G' ? n - 1 : n;
+    for (int i = n - 2; ~i; --i) {
+        if (a[i].second == 'G') {
+            nxt[i] = i;
+        } else {
+            nxt[i] = nxt[i + 1];
+        }
+    }
+    int g = -1;
+    ll res = 0;
+    for (int i = 0; i < n; ++i) {
+        if (a[i].second == 'G') {
+            if (g != -1) {
+                res += a[i].first - g;
+            }
+            g = a[i].first;
+        }
+    }
+        int prevr = -1, prevb = -1;
+        int has_g = 0;
+        int hasr = 0, hasb = 0;
+        vector<int> segsr, segsb;
+    for (int i = 0; i < n; ++i) {
+        if (a[i].second == 'R') {
+            if (prevr != -1) {
+                segsr.emplace_back(a[i].first - a[prevr].first);
+            }
+            prevr = i;
+            hasr = 1;
+        } if (a[i].second == 'B') {
+            if (prevb != -1) {
+                segsb.emplace_back(a[i].first - a[prevb].first);
+            }
+            prevb = i;
+            hasb = 1;
+        } else if (a[i].second == 'G') {
+            if (prevr != -1) {
+                segsr.emplace_back(a[i].first - a[prevr].first);
+            }
+            prevr = i;
+            if (prevb != -1) {
+                segsb.emplace_back(a[i].first - a[prevb].first);
+            }
+            prevb = i;
+            if (has_g) {
+                if (hasr and hasb) {
+                    int len = accumulate(segsb.begin(), segsb.end(), 0);
+                    res += min(len, 2 * len - *max_element(segsr.begin(), segsr.end()) - *max_element(segsb.begin(), segsb.end()));
+                } else if (hasr) {
+                    res += accumulate(segsr.begin(), segsr.end(), 0) - *max_element(segsr.begin(), segsr.end());
+                } else if (hasb) {
+                    res += accumulate(segsb.begin(), segsb.end(), 0) - *max_element(segsb.begin(), segsb.end());
+                }
+            } else {
+                if (hasr) {
+                    res += accumulate(segsr.begin(), segsr.end(), 0);
+                }
+                if (hasb) {
+                    res += accumulate(segsb.begin(), segsb.end(), 0);
+                }
+            }
+            segsr.clear();
+            segsb.clear();
+            has_g = 1;
+            hasr = hasb = 0;
+        }
+    }
+    if (hasr) {
+        res += accumulate(segsr.begin(), segsr.end(), 0);
+    }
+    if (hasb) {
+        res += accumulate(segsb.begin(), segsb.end(), 0);
+    }
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-915f.cc

@@ -0,0 +1,628 @@
+// #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
+#pragma GCC optimize("Ofast,unroll-loops")
+/************* This code requires C++17. ***************/
+
+#include<bits/stdc++.h>
+using namespace std;
+
+/* macro helpers */
+#define __NARGS(...) std::tuple_size<decltype(std::make_tuple(__VA_ARGS__))>::value
+#define __DECOMPOSE_S(a, x) auto x = a;
+#define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
+constexpr void __() {}
+#define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
+#define __as_typeof(container) remove_reference<decltype(container)>::type
+template <typename T> struct argument_type;
+template <typename T, typename U> struct argument_type<T(U)> { using type = U; };
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937_64 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define adj(ch, n) __AS_PROCEDURE(vector<vector<int>> ch((n) + 1);)
+#define edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v), ch[v].push_back(u);)
+#define edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__), ch[v].emplace_back(u, __VA_ARGS__);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__);)
+template <typename T, typename Iterator> pair<size_t, map<T, size_t>> discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    map<T, size_t> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, safe_hash>> unordered_discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    unordered_map<T, size_t, safe_hash> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+
+/* io */
+#define untie __AS_PROCEDURE(ios_base::sync_with_stdio(0), cin.tie(NULL))
+
+// add declarations to avoid cyclic dependency
+template<typename T, typename U> istream& operator>>(istream&, pair<T, U>&);
+template<typename T, typename U> ostream& operator<<(ostream&, const pair<T, U>&);
+template<typename T, size_t N> istream& operator>>(istream&, array<T, N>&);
+template <typename T, size_t N> ostream& operator<<(ostream&, const array<T, N>&);
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>&, const std::tuple<Args...>&);
+template<typename T> ostream& operator<<(ostream&, const vector<T>&);
+std::ostream& operator<<(std::ostream&, const int128&);
+
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename T, size_t N> istream& operator>>(istream& in, array<T, N>& a) {
+    for (size_t i = 0; i < N; ++i) in >> a[i];
+    return in;
+}
+template <typename T, size_t N> ostream& operator<<(ostream& out, const array<T, N>& a) {
+    for (auto&& i : a) out << i << ' ';
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+template<typename T> void __read(T& x) { cin >> x; }
+template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __read(args...); }
+#define read(t, ...) __AS_PROCEDURE(argument_type<void(t)>::type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a(n); for (auto& x : a) cin >> x;)
+#define readvec1(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a((n) + 1); copy_n(ii<argument_type<void(t)>::type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+
+/* pops */
+template <typename Container>
+inline auto poptop(Container& q) {
+    auto ret = q.top();
+    q.pop();
+    return ret;
+}
+template <typename Container>
+inline auto popback(Container& q) {
+    auto ret = q.back();
+    q.pop_back();
+    return ret;
+}
+template <typename Container>
+inline auto popfront(Container& q) {
+    auto ret = q.front();
+    q.pop_front();
+    return ret;
+}
+
+/* math */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
+inline ll root(ll n, ll i) {
+    ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
+    while (l < r) {
+        ll mid = l + r + 1 >> 1;
+        if (qpow<int128>(mid, i) <= n) {
+            l = mid;
+        } else {
+            r = mid - 1;
+        }
+    }
+    return l;
+}
+#pragma GCC diagnostic pop
+
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+#define fastcomb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] * factrev[k] * factrev[(n) - (k)])
+
+__attribute__((target("lzcnt")))
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+template <typename T>
+T mygcd(T a, T b) { return b == 0 ? a : mygcd(b, a % b); }
+
+void __exgcd(ll a, ll b, ll& x, ll& y) {
+  if (b == 0) {
+    x = 1, y = 0;
+    return;
+  }
+  __exgcd(b, a % b, y, x);
+  y -= a / b * x;
+}
+
+ll inverse(ll a, ll b) {
+    ll x, y;
+    __exgcd(a, b, x, y);
+    return mod(x, b);
+}
+
+vector<tuple<int, int, ll>> decompose(ll x) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+    int n = (int)t.length();
+    vector<int> pi(n);
+    for (int i = 1; i < n; i++) {
+        int j = pi[i - 1];
+        while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+        if (t[i] == t[j]) j++;
+        pi[i] = j;
+    }
+    return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+    string cur = t + '#' + s;
+    int sz1 = s.size(), sz2 = t.size();
+    vector<int> v;
+    vector<int> lps = calc_next(cur);
+    for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+        if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+    }
+    return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    MLL& operator+=(const MLL& rhs) { return *this = *this + rhs; }
+    MLL& operator-=(const MLL& rhs) { return *this = *this - rhs; }
+    MLL& operator*=(const MLL& rhs) { return *this = *this * rhs; }
+    MLL& operator/=(const MLL& rhs) { return *this = *this / rhs; }
+    MLL& operator%=(const MLL& rhs) { return *this = *this % rhs; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) ([&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);})
+#define expr(ret, ...) ([&] (__VA_ARGS__) { return (ret); })
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+namespace detail {
+    template <typename T, std::size_t...Is>
+    constexpr std::array<T, sizeof...(Is)>
+    make_array(const T& value, std::index_sequence<Is...>) {
+        return {{(static_cast<void>(Is), value)...}};
+    }
+}
+
+template <typename T, std::size_t N>
+constexpr std::array<T, N> __initarray(const T& value) {
+    return detail::make_array(value, std::make_index_sequence<N>());
+}
+/*******************************************************/
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+class quick_union {
+private:
+    vector<size_t> c, sz;
+public:
+    quick_union(size_t n) : c(n), sz(n) {
+        iota(c.begin(), c.end(), 0);
+        sz.assign(n, 1);
+    }
+    size_t query(size_t i) {
+        if (c[i] != i) c[i] = query(c[i]);
+        return c[i];
+    }
+    void merge(size_t i, size_t j) {
+        if (connected(i, j)) return;
+        sz[query(j)] += sz[query(i)];
+        c[query(i)] = query(j);
+    }
+    bool connected(size_t i, size_t j) {
+        return query(i) == query(j);
+    }
+    size_t query_size(size_t i) {
+        return sz[query(i)];
+    }
+};
+
+// __attribute__((target("popcnt")))
+void solve() {
+    read(int, n);
+    readvec1(int, a, n);
+    adj(ch, n);
+    for (int i = 0; i < n - 1; ++i) {
+        read(int, u, v);
+        edge(ch, u, v);
+    }
+    vector<int> seq(n);
+    iota(seq.begin(), seq.end(), 1);
+    auto work = [&] {
+        ll ret = 0;
+        vector<int> rev(n + 1);
+        for (int i = 0; i < n; ++i) {
+            rev[seq[i]] = i;
+        }
+        quick_union qu(n + 1);
+        for (auto&& v : seq) {
+            ll tot = 0;
+            ll sf = 0;
+            for (auto&& u : ch[v]) {
+                if (rev[u] > rev[v]) continue;
+                tot += qu.query_size(u);
+                sf += qu.query_size(u) * qu.query_size(u);
+                ret += a[v] * qu.query_size(u);
+                qu.merge(u, v);
+            }
+            ret += a[v];
+            ret += a[v] * (tot * tot - sf) / 2;
+        }
+        return ret;
+    };
+    ll res = 0;
+    sort_by_key(seq.begin(), seq.end(), expr(a[v], int v));
+    res += work();
+    sort_by_key(seq.begin(), seq.end(), expr(-a[v], int v));
+    res -= work();
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-915g.cc

@@ -0,0 +1,612 @@
+// #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
+#pragma GCC optimize("Ofast,unroll-loops")
+/************* This code requires C++17. ***************/
+
+#include<bits/stdc++.h>
+using namespace std;
+
+/* macro helpers */
+#define __NARGS(...) std::tuple_size<decltype(std::make_tuple(__VA_ARGS__))>::value
+#define __DECOMPOSE_S(a, x) auto x = a;
+#define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
+constexpr void __() {}
+#define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
+#define __as_typeof(container) remove_reference<decltype(container)>::type
+template <typename T> struct argument_type;
+template <typename T, typename U> struct argument_type<T(U)> { using type = U; };
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937_64 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define adj(ch, n) __AS_PROCEDURE(vector<vector<int>> ch((n) + 1);)
+#define edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v), ch[v].push_back(u);)
+#define edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__), ch[v].emplace_back(u, __VA_ARGS__);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__);)
+template <typename T, typename Iterator> pair<size_t, map<T, size_t>> discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    map<T, size_t> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, safe_hash>> unordered_discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    unordered_map<T, size_t, safe_hash> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+
+/* io */
+#define untie __AS_PROCEDURE(ios_base::sync_with_stdio(0), cin.tie(NULL))
+
+// add declarations to avoid cyclic dependency
+template<typename T, typename U> istream& operator>>(istream&, pair<T, U>&);
+template<typename T, typename U> ostream& operator<<(ostream&, const pair<T, U>&);
+template<typename T, size_t N> istream& operator>>(istream&, array<T, N>&);
+template <typename T, size_t N> ostream& operator<<(ostream&, const array<T, N>&);
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>&, const std::tuple<Args...>&);
+template<typename T> ostream& operator<<(ostream&, const vector<T>&);
+std::ostream& operator<<(std::ostream&, const int128&);
+
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename T, size_t N> istream& operator>>(istream& in, array<T, N>& a) {
+    for (size_t i = 0; i < N; ++i) in >> a[i];
+    return in;
+}
+template <typename T, size_t N> ostream& operator<<(ostream& out, const array<T, N>& a) {
+    for (auto&& i : a) out << i << ' ';
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+template<typename T> void __read(T& x) { cin >> x; }
+template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __read(args...); }
+#define read(t, ...) __AS_PROCEDURE(argument_type<void(t)>::type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a(n); for (auto& x : a) cin >> x;)
+#define readvec1(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a((n) + 1); copy_n(ii<argument_type<void(t)>::type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+
+/* pops */
+template <typename Container>
+inline auto poptop(Container& q) {
+    auto ret = q.top();
+    q.pop();
+    return ret;
+}
+template <typename Container>
+inline auto popback(Container& q) {
+    auto ret = q.back();
+    q.pop_back();
+    return ret;
+}
+template <typename Container>
+inline auto popfront(Container& q) {
+    auto ret = q.front();
+    q.pop_front();
+    return ret;
+}
+
+/* math */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
+inline ll root(ll n, ll i) {
+    ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
+    while (l < r) {
+        ll mid = l + r + 1 >> 1;
+        if (qpow<int128>(mid, i) <= n) {
+            l = mid;
+        } else {
+            r = mid - 1;
+        }
+    }
+    return l;
+}
+#pragma GCC diagnostic pop
+
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+#define fastcomb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] * factrev[k] * factrev[(n) - (k)])
+
+__attribute__((target("lzcnt")))
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+template <typename T>
+T mygcd(T a, T b) { return b == 0 ? a : mygcd(b, a % b); }
+
+void __exgcd(ll a, ll b, ll& x, ll& y) {
+  if (b == 0) {
+    x = 1, y = 0;
+    return;
+  }
+  __exgcd(b, a % b, y, x);
+  y -= a / b * x;
+}
+
+ll inverse(ll a, ll b) {
+    ll x, y;
+    __exgcd(a, b, x, y);
+    return mod(x, b);
+}
+
+vector<tuple<int, int, ll>> decompose(ll x) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+    int n = (int)t.length();
+    vector<int> pi(n);
+    for (int i = 1; i < n; i++) {
+        int j = pi[i - 1];
+        while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+        if (t[i] == t[j]) j++;
+        pi[i] = j;
+    }
+    return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+    string cur = t + '#' + s;
+    int sz1 = s.size(), sz2 = t.size();
+    vector<int> v;
+    vector<int> lps = calc_next(cur);
+    for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+        if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+    }
+    return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    MLL& operator+=(const MLL& rhs) { return *this = *this + rhs; }
+    MLL& operator-=(const MLL& rhs) { return *this = *this - rhs; }
+    MLL& operator*=(const MLL& rhs) { return *this = *this * rhs; }
+    MLL& operator/=(const MLL& rhs) { return *this = *this / rhs; }
+    MLL& operator%=(const MLL& rhs) { return *this = *this % rhs; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) ([&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);})
+#define expr(ret, ...) ([&] (__VA_ARGS__) { return (ret); })
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+namespace detail {
+    template <typename T, std::size_t...Is>
+    constexpr std::array<T, sizeof...(Is)>
+    make_array(const T& value, std::index_sequence<Is...>) {
+        return {{(static_cast<void>(Is), value)...}};
+    }
+}
+
+template <typename T, std::size_t N>
+constexpr std::array<T, N> __initarray(const T& value) {
+    return detail::make_array(value, std::make_index_sequence<N>());
+}
+/*******************************************************/
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+constexpr int N = 2e6 + 10;
+int mu[N];
+
+void get_mu(int n) {
+    vector<bool> not_prime(n + 1);
+    vector<int> res;
+    mu[1] = 1;
+    for (int i = 2; i <= n; ++i) {
+        if (not not_prime[i]) {
+            res.emplace_back(i);
+            mu[i] = -1;
+        }
+        for (auto&& x : res) {
+            if (i * x > n) break;
+            not_prime[i * x] = 1;
+            if (i % x == 0) {
+                mu[i * x] = 0;
+                break;
+            } else {
+                mu[i * x] = -mu[i];
+            }
+        }
+    }
+}
+
+void prep() {
+    get_mu(N - 1);
+}
+
+// __attribute__((target("popcnt")))
+void solve() {
+    using mll = MLL<MDL>;
+    read(int, n, k);
+    vector<mll> pw(k + 1);
+    for (int i = 1; i <= k; ++i) {
+        pw[i] = qpow(i, n, MDL);
+    }
+    vector<mll> diff(k + 2);
+    for (int i = 1; i <= k; ++i) {
+        for (int j = i; j <= k; j += i) {
+            diff[j] += mu[i] * pw[j / i];
+            diff[min(k + 1, j + i)] -= mu[i] * pw[j / i];
+        }
+    }
+    mll curr = 0;
+    mll res = 0;
+    for (int i = 1; i <= k; ++i) {
+        curr += diff[i];
+        res += curr.val xor i;
+    }
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-960e.cc

@@ -0,0 +1,611 @@
+// #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
+#pragma GCC optimize("Ofast")
+/************* This code requires C++17. ***************/
+
+#include<bits/stdc++.h>
+using namespace std;
+
+/* macro helpers */
+#define __NARGS(...) std::tuple_size<decltype(std::make_tuple(__VA_ARGS__))>::value
+#define __DECOMPOSE_S(a, x) auto x = a;
+#define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
+constexpr void __() {}
+#define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
+#define __as_typeof(container) remove_reference<decltype(container)>::type
+template <typename T> struct argument_type;
+template <typename T, typename U> struct argument_type<T(U)> { using type = U; };
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937_64 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define adj(ch, n) __AS_PROCEDURE(vector<vector<int>> ch((n) + 1);)
+#define edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v), ch[v].push_back(u);)
+#define edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__), ch[v].emplace_back(u, __VA_ARGS__);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__);)
+template <typename T, typename Iterator> pair<size_t, map<T, size_t>> discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    map<T, size_t> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, safe_hash>> unordered_discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    unordered_map<T, size_t, safe_hash> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+
+/* io */
+#define untie __AS_PROCEDURE(ios_base::sync_with_stdio(0), cin.tie(NULL))
+
+// add declarations to avoid cyclic dependency
+template<typename T, typename U> istream& operator>>(istream&, pair<T, U>&);
+template<typename T, typename U> ostream& operator<<(ostream&, const pair<T, U>&);
+template<typename T, size_t N> istream& operator>>(istream&, array<T, N>&);
+template <typename T, size_t N> ostream& operator<<(ostream&, const array<T, N>&);
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>&, const std::tuple<Args...>&);
+template<typename T> ostream& operator<<(ostream&, const vector<T>&);
+std::ostream& operator<<(std::ostream&, const int128&);
+
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename T, size_t N> istream& operator>>(istream& in, array<T, N>& a) {
+    for (size_t i = 0; i < N; ++i) in >> a[i];
+    return in;
+}
+template <typename T, size_t N> ostream& operator<<(ostream& out, const array<T, N>& a) {
+    for (auto&& i : a) out << i << ' ';
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+template<typename T> void __read(T& x) { cin >> x; }
+template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __read(args...); }
+#define read(t, ...) __AS_PROCEDURE(argument_type<void(t)>::type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a(n); for (auto& x : a) cin >> x;)
+#define readvec1(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a((n) + 1); copy_n(ii<argument_type<void(t)>::type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+
+/* pops */
+template <typename Container>
+inline auto poptop(Container& q) {
+    auto ret = q.top();
+    q.pop();
+    return ret;
+}
+template <typename Container>
+inline auto popback(Container& q) {
+    auto ret = q.back();
+    q.pop_back();
+    return ret;
+}
+template <typename Container>
+inline auto popfront(Container& q) {
+    auto ret = q.front();
+    q.pop_front();
+    return ret;
+}
+
+/* math */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
+inline ll root(ll n, ll i) {
+    ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
+    while (l < r) {
+        ll mid = l + r + 1 >> 1;
+        if (qpow<int128>(mid, i) <= n) {
+            l = mid;
+        } else {
+            r = mid - 1;
+        }
+    }
+    return l;
+}
+#pragma GCC diagnostic pop
+
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+#define fastcomb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] * factrev[k] * factrev[(n) - (k)])
+
+__attribute__((target("lzcnt")))
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+template <typename T>
+T mygcd(T a, T b) { return b == 0 ? a : mygcd(b, a % b); }
+
+void __exgcd(ll a, ll b, ll& x, ll& y) {
+  if (b == 0) {
+    x = 1, y = 0;
+    return;
+  }
+  __exgcd(b, a % b, y, x);
+  y -= a / b * x;
+}
+
+ll inverse(ll a, ll b) {
+    ll x, y;
+    __exgcd(a, b, x, y);
+    return mod(x, b);
+}
+
+vector<tuple<int, int, ll>> decompose(ll x) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+    int n = (int)t.length();
+    vector<int> pi(n);
+    for (int i = 1; i < n; i++) {
+        int j = pi[i - 1];
+        while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+        if (t[i] == t[j]) j++;
+        pi[i] = j;
+    }
+    return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+    string cur = t + '#' + s;
+    int sz1 = s.size(), sz2 = t.size();
+    vector<int> v;
+    vector<int> lps = calc_next(cur);
+    for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+        if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+    }
+    return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    MLL& operator+=(const MLL& rhs) { return *this = *this + rhs; }
+    MLL& operator-=(const MLL& rhs) { return *this = *this - rhs; }
+    MLL& operator*=(const MLL& rhs) { return *this = *this * rhs; }
+    MLL& operator/=(const MLL& rhs) { return *this = *this / rhs; }
+    MLL& operator%=(const MLL& rhs) { return *this = *this % rhs; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) ([&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);})
+#define expr(ret, ...) ([&] (__VA_ARGS__) { return (ret); })
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+namespace detail {
+    template <typename T, std::size_t...Is>
+    constexpr std::array<T, sizeof...(Is)>
+    make_array(const T& value, std::index_sequence<Is...>) {
+        return {{(static_cast<void>(Is), value)...}};
+    }
+}
+
+template <typename T, std::size_t N>
+constexpr std::array<T, N> __initarray(const T& value) {
+    return detail::make_array(value, std::make_index_sequence<N>());
+}
+/*******************************************************/
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+// __attribute__((target("popcnt")))
+void solve() {
+    using mll = MLL<MDL>;
+
+    read(int, n);
+    readvec1(int, a, n);
+    adj(ch, n);
+    for (int i = 0; i < n - 1; ++i) {
+        read(int, u, v);
+        edge(ch, u, v);
+    }
+    vector<array<mll, 2>> dp(n + 1);
+    vector<int> sz(n + 1);
+    {
+        auto dfs = [&] (auto dfs, int v, int pa) -> void {
+            dp[v][1] = 1;
+            sz[v] = 1;
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                dfs(dfs, u, v);
+                dp[v][0] += dp[u][1];
+                dp[v][1] += dp[u][0];
+                sz[v] += sz[u];
+            }
+        };
+        dfs(dfs, 1, 0);
+    }
+    mll res = 0;
+    {
+        auto dfs = [&] (auto dfs, int v, int pa, array<mll, 2> up) -> void {
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                res += (dp[u][0] - dp[u][1]) * (n - sz[u]) * a[v];
+            }
+            res += (up[1] - up[0]) * sz[v] * a[v];
+            res += mll(n) * a[v];
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                dfs(dfs, u, v, {
+                    up[1] + dp[v][1] - dp[u][0],
+                    up[0] + dp[v][0] - dp[u][1],
+                });
+            }
+        };
+        dfs(dfs, 1, 0, { 0, 0 });
+    }
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-965e.cc

@@ -0,0 +1,751 @@
+// #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
+#pragma GCC optimize("Ofast")
+/************* This code requires C++17. ***************/
+
+#include<bits/stdc++.h>
+using namespace std;
+
+/* macro helpers */
+#define __NARGS(...) std::tuple_size<decltype(std::make_tuple(__VA_ARGS__))>::value
+#define __DECOMPOSE_S(a, x) auto x = a;
+#define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
+constexpr void __() {}
+#define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
+#define __as_typeof(container) remove_reference<decltype(container)>::type
+template <typename T> struct argument_type;
+template <typename T, typename U> struct argument_type<T(U)> { using type = U; };
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937_64 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define adj(ch, n) __AS_PROCEDURE(vector<vector<int>> ch((n) + 1);)
+#define edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v), ch[v].push_back(u);)
+#define edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__), ch[v].emplace_back(u, __VA_ARGS__);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, ...) __AS_PROCEDURE(ch[u].emplace_back(v, __VA_ARGS__);)
+template <typename T, typename Iterator> pair<size_t, map<T, size_t>> discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    map<T, size_t> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, safe_hash>> unordered_discretize(Iterator __first, Iterator __last) {
+    set<T> st(__first, __last);
+    size_t N = 0;
+    unordered_map<T, size_t, safe_hash> mp;
+    for (auto&& x : st) mp[x] = ++N;
+    return {N, mp};
+}
+
+/* io */
+#define untie __AS_PROCEDURE(ios_base::sync_with_stdio(0), cin.tie(NULL))
+
+// add declarations to avoid cyclic dependency
+template<typename T, typename U> istream& operator>>(istream&, pair<T, U>&);
+template<typename T, typename U> ostream& operator<<(ostream&, const pair<T, U>&);
+template<typename T, size_t N> istream& operator>>(istream&, array<T, N>&);
+template <typename T, size_t N> ostream& operator<<(ostream&, const array<T, N>&);
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>&, const std::tuple<Args...>&);
+template<typename T> ostream& operator<<(ostream&, const vector<T>&);
+std::ostream& operator<<(std::ostream&, const int128&);
+
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename T, size_t N> istream& operator>>(istream& in, array<T, N>& a) {
+    for (size_t i = 0; i < N; ++i) in >> a[i];
+    return in;
+}
+template <typename T, size_t N> ostream& operator<<(ostream& out, const array<T, N>& a) {
+    for (auto&& i : a) out << i << ' ';
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+template<typename T> void __read(T& x) { cin >> x; }
+template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __read(args...); }
+#define read(t, ...) __AS_PROCEDURE(argument_type<void(t)>::type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a(n); for (auto& x : a) cin >> x;)
+#define readvec1(t, a, n) __AS_PROCEDURE(vector<argument_type<void(t)>::type> a((n) + 1); copy_n(ii<argument_type<void(t)>::type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+
+/* pops */
+template <typename Container>
+inline auto poptop(Container& q) {
+    auto ret = q.top();
+    q.pop();
+    return ret;
+}
+template <typename Container>
+inline auto popback(Container& q) {
+    auto ret = q.back();
+    q.pop_back();
+    return ret;
+}
+template <typename Container>
+inline auto popfront(Container& q) {
+    auto ret = q.front();
+    q.pop_front();
+    return ret;
+}
+
+/* math */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
+inline ll root(ll n, ll i) {
+    ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
+    while (l < r) {
+        ll mid = l + r + 1 >> 1;
+        if (qpow<int128>(mid, i) <= n) {
+            l = mid;
+        } else {
+            r = mid - 1;
+        }
+    }
+    return l;
+}
+#pragma GCC diagnostic pop
+
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+#define fastcomb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] * factrev[k] * factrev[(n) - (k)])
+
+__attribute__((target("lzcnt")))
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+template <typename T>
+T mygcd(T a, T b) { return b == 0 ? a : mygcd(b, a % b); }
+
+void __exgcd(ll a, ll b, ll& x, ll& y) {
+  if (b == 0) {
+    x = 1, y = 0;
+    return;
+  }
+  __exgcd(b, a % b, y, x);
+  y -= a / b * x;
+}
+
+ll inverse(ll a, ll b) {
+    ll x, y;
+    __exgcd(a, b, x, y);
+    return mod(x, b);
+}
+
+vector<tuple<int, int, ll>> decompose(ll x) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+    int n = (int)t.length();
+    vector<int> pi(n);
+    for (int i = 1; i < n; i++) {
+        int j = pi[i - 1];
+        while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+        if (t[i] == t[j]) j++;
+        pi[i] = j;
+    }
+    return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+    string cur = t + '#' + s;
+    int sz1 = s.size(), sz2 = t.size();
+    vector<int> v;
+    vector<int> lps = calc_next(cur);
+    for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+        if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+    }
+    return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    MLL& operator+=(const MLL& rhs) { return *this = *this + rhs; }
+    MLL& operator-=(const MLL& rhs) { return *this = *this - rhs; }
+    MLL& operator*=(const MLL& rhs) { return *this = *this * rhs; }
+    MLL& operator/=(const MLL& rhs) { return *this = *this / rhs; }
+    MLL& operator%=(const MLL& rhs) { return *this = *this % rhs; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) ([&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);})
+#define expr(ret, ...) ([&] (__VA_ARGS__) { return (ret); })
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+namespace detail {
+    template <typename T, std::size_t...Is>
+    constexpr std::array<T, sizeof...(Is)>
+    make_array(const T& value, std::index_sequence<Is...>) {
+        return {{(static_cast<void>(Is), value)...}};
+    }
+}
+
+template <typename T, std::size_t N>
+constexpr std::array<T, N> __initarray(const T& value) {
+    return detail::make_array(value, std::make_index_sequence<N>());
+}
+/*******************************************************/
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+template<typename Addable_Info_t, typename Tag_t, typename Sequence = std::vector<Addable_Info_t>> class segtree {
+private:
+    using size_type = uint64_t;
+    using info_type = Addable_Info_t;
+    using tag_type = Tag_t;
+    size_type _max;
+    vector<info_type> d;
+    vector<tag_type> b;
+    void pull(size_type p) {
+        d[p] = d[p * 2] + d[p * 2 + 1];
+    }
+    void push(size_type p, size_type left_len, size_type right_len) {
+        d[p * 2].apply(b[p], left_len), d[p * 2 + 1].apply(b[p], right_len);
+        b[p * 2].apply(b[p]), b[p * 2 + 1].apply(b[p]);
+        b[p] = tag_type();
+    }
+    void set(size_type s, size_type t, size_type p, size_type x, const info_type& c) {
+        if (s == t) {
+            d[p] = c;
+            return;
+        }
+        size_type m = s + (t - s >> 1);
+        if (s != t) push(p, m - s + 1, t - m);
+        if (x <= m) set(s, m, p * 2, x, c);
+        else set(m + 1, t, p * 2 + 1, x, c);
+        pull(p);
+    }
+
+    void range_apply(size_type s, size_type t, size_type p, size_type l, size_type r, const tag_type& c) {
+        if (l <= s && t <= r) {
+            d[p].apply(c, t - s + 1);
+            b[p].apply(c);
+            return;
+        }
+        size_type m = s + (t - s >> 1);
+        push(p, m - s + 1, t - m);
+        if (l <= m) range_apply(s, m, p * 2, l, r, c);
+        if (r > m)  range_apply(m + 1, t, p * 2 + 1, l, r, c);
+        pull(p);
+    }
+    info_type range_query(size_type s, size_type t, size_type p, size_type l, size_type r) {
+        if (l <= s && t <= r) {
+            return d[p];
+        }
+        size_type m = s + (t - s >> 1);
+        info_type res = {};
+        push(p, m - s + 1, t - m);
+        if (l <= m) res = res + range_query(s, m, p * 2, l, r);
+        if (r > m)  res = res + range_query(m + 1, t, p * 2 + 1, l, r);
+        return res;
+    }
+    void build(const Sequence& a, size_type s, size_type t, size_type p) {
+        if (s == t) {
+            d[p] = a[s];
+            return;
+        }
+        int m = s + (t - s >> 1);
+        build(a, s, m, p * 2);
+        build(a, m + 1, t, p * 2 + 1);
+        pull(p);
+    }
+public:
+    segtree(size_type __max) : d(4 * __max), b(4 * __max), _max(__max - 1) {}
+    segtree(const Sequence& a) : segtree(a.size()) {
+        build(a, {}, _max, 1);
+    }
+    void set(size_type i, const info_type& c) {
+        set({}, _max, 1, i, c);
+    }
+
+    void range_apply(size_type l, size_type r, const tag_type& c) {
+        range_apply({}, _max, 1, l, r, c);
+    }
+    void apply(size_type i, const tag_type& c) {
+        range_apply(i, i, c);
+    }
+    info_type range_query(size_type l, size_type r) {
+        return range_query({}, _max, 1, l, r);
+    }
+    info_type query(size_type i) {
+        return range_query(i, i);
+    }
+    Sequence serialize() {
+        Sequence res = {};
+        for (size_type i = 0; i <= _max; ++i) {
+            res.push_back(query(i));
+        }
+        return res;
+    }
+    const vector<info_type>& get_d() {
+        return d;
+    }
+};
+struct Tag {
+    void apply(const Tag& rhs) { }
+};
+struct Info {
+    int val = -INF;
+    int pos = 0;
+    void apply(const Tag& rhs, size_t len) { }
+};
+Info operator+(const Info &a, const Info &b) {
+    return {
+        .val = max(a.val, b.val),
+        .pos = a.val > b.val ? a.pos : b.pos,
+    };
+}
+
+// __attribute__((target("popcnt")))
+void solve() {
+    read(int, m);
+    readvec(string, a, m);
+    vector<pair<array<int, 26>, bool>> trie(1);
+    auto insert = [&] (const string& s) -> void {
+        int curr = 0;
+        for (auto c : s) {
+            c -= 'a';
+            if (not trie[curr].first[c]) {
+                trie[curr].first[c] = trie.size();
+                trie.emplace_back();
+            }
+            curr = trie[curr].first[c];
+        }
+        trie[curr].second = 1;
+    };
+    for_each(a.begin(), a.end(), insert);
+    int n = trie.size();
+    vector<int> dfn(n);
+    vector<int> depth(n);
+    depth[0] = -1;
+    vector<int> sz(n);
+    segtree<Info, Tag> tr(n);
+    {
+        int t = 0;
+        auto dfs = [&] (auto dfs, int v, int pa) -> void {
+            depth[v] = depth[pa] + 1;
+            sz[v] = 1;
+            dfn[v] = t++;
+            if (trie[v].second) {
+                tr.set(dfn[v], {
+                    .val = depth[v],
+                    .pos = v,
+                });
+            }
+            for (auto&& u : trie[v].first) {
+                if (u != 0) {
+                    dfs(dfs, u, v);
+                    sz[v] += sz[u];
+                }
+            }
+        };
+        dfs(dfs, 0, 0);
+    }
+    {
+        auto dfs = [&] (auto dfs, int v) -> void {
+            for (auto&& u : trie[v].first) {
+                if (u != 0) {
+                    dfs(dfs, u);
+                }
+            }
+            if (v != 0 and not trie[v].second) {
+                auto [val, u] = tr.range_query(dfn[v], dfn[v] + sz[v] - 1);
+                if (val != -INF) {
+                    tr.set(dfn[u], { -INF, 0 });
+                    tr.set(dfn[v], { depth[v], v });
+                    trie[u].second = 0;
+                    trie[v].second = 1;
+                }
+            }
+        };
+        dfs(dfs, 0);
+    }
+    int res = 0;
+    {
+        auto dfs = [&] (auto dfs, int v) -> void {
+            if (trie[v].second) {
+                res += depth[v];
+            }
+            for (auto&& u : trie[v].first) {
+                if (u != 0) {
+                    dfs(dfs, u);
+                }
+            }
+        };
+        dfs(dfs, 0);
+    }
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/d.cc

@@ -1,5 +1,5 @@
 // #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
-#pragma GCC optimize("Ofast")
+#pragma GCC optimize("Ofast,unroll-loops")
 /************* This code requires C++17. ***************/
 
 #include<bits/stdc++.h>
@@ -279,9 +279,19 @@ return_t qpow(ll b, ll p) {
     else return half * half;
 }
 
-// Accurately find `i` 'th root of `n` (taking the floor)
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wparentheses"
+// Accurately find `i` 'th root of `n` (taking the floor)
 inline ll root(ll n, ll i) {
     ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
     while (l < r) {
@@ -522,13 +532,68 @@ void dump() {}
 
 void dump_ignore() {}
 
+constexpr int N = 4e5 + 10;
+bool not_prime[N + 1];
+
+void soe(int n) {
+    vector<int> res;
+    for (int i = 2; i <= n; ++i) {
+        if (not not_prime[i]) {
+            res.emplace_back(i);
+        }
+        for (auto&& x : res) {
+            if (i * x > n) break;
+            not_prime[i * x] = 1;
+            if (i % x == 0) break;
+        }
+    }
+}
+
 void prep() {
+    soe(N);
 }
 
 // __attribute__((target("popcnt")))
 void solve() {
-    read(int, n, m);
+    read(int, n);
-    
+    adj(ch, n);
+    for (int i = 0; i < n - 1; ++i) {
+        read(int, u, v);
+        edge(ch, u, v);
+    }
+    array<vector<int>, 2> bk;
+    vector<int> fa(n + 1);
+    {
+        auto dfs = [&] (auto dfs, int v, int pa, int par) -> void {
+            fa[v] = pa;
+            bk[par].emplace_back(v);
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                dfs(dfs, u, v, 1 - par);
+            }
+        };
+        dfs(dfs, 1, 0, 0);
+    }
+    vector<int> res(n + 1);
+    int curr = 2;
+    for (auto&& v : bk[0]) {
+        res[v] = curr;
+        curr += 2;
+    }
+    curr = 2 * n;
+    for (auto&& v : bk[1]) {
+        res[v] = curr;
+        curr -= 2;
+    }
+    for (int i = 2; i <= n; ++i) {
+        if (abs(res[fa[i]] - res[i]) == 2) {
+            res[i] = res[fa[i]] > res[i] ? res[i] + 1 : res[i] - 1;
+            break;
+        }
+    }
+    for (int i = 1; i <= n; ++i) {
+        cout << res[i] << " \n"[i == n];
+    }
 }
 
 int main() {

src/bin/template.cc

@@ -1,5 +1,5 @@
 // #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
-#pragma GCC optimize("Ofast")
+#pragma GCC optimize("Ofast,unroll-loops")
 /************* This code requires C++17. ***************/
 
 #include<bits/stdc++.h>

src/bin/test.cc

@@ -1,5 +1,5 @@
 // #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
-#pragma GCC optimize("Ofast")
+#pragma GCC optimize("Ofast,unroll-loops")
 /************* This code requires C++17. ***************/
 
 #include<bits/stdc++.h>
@@ -172,6 +172,17 @@ template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, s
 
 /* io */
 #define untie __AS_PROCEDURE(ios_base::sync_with_stdio(0), cin.tie(NULL))
+
+// add declarations to avoid cyclic dependency
+template<typename T, typename U> istream& operator>>(istream&, pair<T, U>&);
+template<typename T, typename U> ostream& operator<<(ostream&, const pair<T, U>&);
+template<typename T, size_t N> istream& operator>>(istream&, array<T, N>&);
+template <typename T, size_t N> ostream& operator<<(ostream&, const array<T, N>&);
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>&, const std::tuple<Args...>&);
+template<typename T> ostream& operator<<(ostream&, const vector<T>&);
+std::ostream& operator<<(std::ostream&, const int128&);
+
 template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
     return in >> p.first >> p.second;
 }
@@ -268,6 +279,18 @@ return_t qpow(ll b, ll p) {
     else return half * half;
 }
 
+// dynamic modulus
+ll qpow(ll b, ll p, ll mod) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    ll half = qpow(b, p / 2, mod);
+    if (p % 2 == 1) return (int128(half) * half % mod) * b % mod;
+    else return half * half % mod;
+}
+
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wparentheses"
 // Accurately find `i` 'th root of `n` (taking the floor)
 inline ll root(ll n, ll i) {
     ll l = 0, r = pow(LLONG_MAX, ld(1) / i);
@@ -281,11 +304,13 @@ inline ll root(ll n, ll i) {
     }
     return l;
 }
+#pragma GCC diagnostic pop
 
 
 #define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
 #define fastcomb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] * factrev[k] * factrev[(n) - (k)])
 
+__attribute__((target("lzcnt")))
 constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
 
 template <typename T>
@@ -510,12 +535,22 @@ void dump_ignore() {}
 void prep() {
 }
 
+template <typename T>
+struct my_any {
+    bool has;
+    T val;
+    my_any(const T& val) : has(true), val(val) {}
+    bool has_value() const { return has; }
+    T& value() { return val; }
+    type_index type() { return typeid(val); }
+};
+
 // __attribute__((target("popcnt")))
 void solve() {
-    constexpr int N = 10;
+    my_any a = 10;
-    [] {
+    my_any b = string("hello");
-        bitset<N + 1> test;
+    assert(a.type() == typeid(int));
-    } ();
+    assert(b.type() == typeid(string));
 }
 
 int main() {