regular backup

src/bin/a.cc

@@ -540,13 +540,29 @@ void prep() {
 
 // __attribute__((target("popcnt")))
 void solve() {
-    read(int, n);
+    int n;
-    if (n % 2 == 0) {
+    unordered_set<string> oc;
-        cout << (n / 2 + 1) * (n / 2 + 1) << '\n';
+    while (1) {
-    } else {
+        read(string, s);
-        int base = ((n - 1) / 2 + 1);
+        if (s[0] >= '0' and s[0] <= '9') {
-        cout << 4 * base * base - (base - 1) * base * 2 << '\n';
+            n = stoi(s);
+            break;
         }
+        string nw;
+        for (auto&& c : s) {
+            if (c >= 'A' and c <= 'Z') {
+                nw += c - 'A' + 'a';
+            } else if (c >= 'a' and c <= 'z') {
+                nw += c;
+            }
+        }
+        oc.emplace(nw);
+    }
+    while (n--) {
+        read(string, s);
+        oc.erase(s);
+    }
+    cout << oc.size() << '\n';
 }
 
 int main() {

src/bin/b.cc

@@ -337,7 +337,7 @@ ll inverse(ll a, ll b) {
 vector<tuple<int, int, ll>> decompose(ll x) {
     // return (factor, count, factor ** count)
     vector<tuple<int, int, ll>> res;
-    for (int i = 2; ll(1) * i * i <= x; i++) {
+    for (int i = 2; i * i <= x; i++) {
         if (x % i == 0) {
             int cnt = 0;
             ll pw = 1;
@@ -527,7 +527,7 @@ constexpr std::array<T, N> __initarray(const T& value) {
 }
 /*******************************************************/
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -540,26 +540,9 @@ void prep() {
 
 // __attribute__((target("popcnt")))
 void solve() {
-    read(ll, p, q);
+    read(string, s);
-    if (p % q == 0) {
+    string s1(s.rbegin(), s.rend());
-        auto f = decompose(q);
+    cout << s1 << '\n';
-        ll pp = p;
-        ll res = INFLL;
-        for (auto&& [x, c, t] : f) {
-            deb(x, c, t);
-            ll pw = 1;
-            while (pp % x == 0) {
-                pp /= x;
-                pw *= x;
-            }
-            assert(pw % t == 0);
-            chmin(res, pw / t * x);
-        }
-        assert(gcd(pp, q) == 1);
-        cout << p / res << '\n';
-    } else {
-        cout << p << '\n';
-    }
 }
 
 int main() {

src/bin/c.cc

@@ -527,7 +527,7 @@ constexpr std::array<T, N> __initarray(const T& value) {
 }
 /*******************************************************/
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -540,25 +540,48 @@ void prep() {
 
 // __attribute__((target("popcnt")))
 void solve() {
-    using mll = MLL<MDL>;
+    read(int, n);
-    read(int, n, k);
+    vector<int> a(n + 1);
-    vector dp(k + 1, vector<mll>(n + 2));
+    a[0] = 2200;
-    vector dp1(k + 1, vector<mll>(n + 2));
+    adj(ch, n);
-    for (int i = 1; i <= k; ++i) {
+    for (int i = 1; i <= n; ++i) {
-        dp[i][n + 1] = 1;
+        read(int, f, x);
-        dp1[i][n + 1] = 1;
+        edge(ch, i, f);
-        for (int j = n; j; --j) {
+        a[i] = x;
-            dp[i][j] = (i > 1) + dp[i][j + 1];
+    }
-            dp1[i][j] = dp[i][j + 1];
+    vector<ll> ps(n + 1);
-            if (i - 2 >= 1) {
+    vector<bool> valid(n + 1);
-                dp[i][j] += dp1[i - 2][j - 1];
+    {
+        auto dfs = [&] (auto dfs, int v, int pa) -> void {
+            int son = 0;
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                son += 1;
+                dfs(dfs, u, v);
+                ps[v] += ps[u];
+            }
+            if (not son) {
+                ps[v] = a[v];
+            } else {
+                valid[v] = 1;
+            }
+        };
+        dfs(dfs, 0, 0);
+    }
+    if (ps[0] > a[0]) {
+        cout << "NO\n";
+        return;
+    }
+
+    sort(a.begin() + 1, a.end());
+    sort(ps.begin() + 1, ps.end());
+    for (int i = 1; i <= n; ++i) {
+        if (ps[i] > a[i]) {
+            cout << "NO\n";
+            return;
         }
     }
-        for (int j = 1; j <= n; ++j) {
+    cout << "YES\n";
-            dp1[i][j] += dp1[i][j - 1];
-        }
-    }
-    cout << dp[k][1] << '\n';
 }
 
 int main() {

src/bin/cf-799e.cc

@@ -0,0 +1,794 @@
+// #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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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);
+    }
+
+    int binary_search(size_type s, size_type t, size_type p, int tot) {
+        if (s == t) {
+            return s;
+        }
+        size_type m = s + (t - s >> 1);
+        if (d[p * 2].cnt >= tot) return binary_search(s, m, p * 2, tot);
+        else return binary_search(m + 1, t, p * 2 + 1, tot - d[p * 2].cnt);
+    }
+
+    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);
+    }
+
+    int binary_search(int cnt) {
+        return binary_search({}, _max, 1, cnt);
+    }
+
+    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 {
+    ll val = 0;
+    int cnt = 0;
+    void apply(const Tag& rhs) {
+        val += rhs.val;
+        cnt += rhs.cnt;
+    }
+};
+
+struct Info {
+    ll val = 0;
+    int cnt = 0;
+    void apply(const Tag& rhs, size_t len) {
+        val += rhs.val;
+        cnt += rhs.cnt;
+    }
+};
+
+Info operator+(const Info &a, const Info &b) {
+    return {
+        .val = a.val + b.val,
+        .cnt = a.cnt + b.cnt,
+    };
+}
+
+
+// __attribute__((target("popcnt")))
+void solve() {
+    read(int, n, m, k);
+    readvec(int, a, n);
+    read(int, p);
+    readvec(int, x, p);
+    read(int, q);
+    readvec(int, y, q);
+
+    transform(x.begin(), x.end(), x.begin(), expr(x - 1, auto x));
+    transform(y.begin(), y.end(), y.begin(), expr(y - 1, auto y));
+
+    vector<int> both, left, right;
+    vector<int> mark(n);
+    for (auto&& i : x) {
+        mark[i] += 1;
+    }
+    for (auto&& i : y) {
+        mark[i] += 1;
+    }
+    ll left_sum = 0, right_sum = 0;
+    for (auto&& i : x) {
+        if (mark[i] == 2) {
+            both.emplace_back(i);
+        } else {
+            left.emplace_back(i);
+            left_sum += a[i];
+        }
+    }
+    for (auto&& i : y) {
+        if (mark[i] == 1) {
+            right.emplace_back(i);
+            right_sum += a[i];
+        }
+    }
+
+    sort_by_key(both.begin(), both.end(), expr(a[i], auto i));
+    sort_by_key(left.begin(), left.end(), expr(a[i], auto i));
+    sort_by_key(right.begin(), right.end(), expr(a[i], auto i));
+
+    vector<int> oc(a.begin(), a.end());
+    sort(oc.begin(), oc.end());
+    int t = unique(oc.begin(), oc.end()) - oc.begin();
+    oc.resize(t);
+    auto get = [&] (int x) { return lower_bound(oc.begin(), oc.end(), x) - oc.begin(); };
+
+    ll res = INFLL;
+
+    segtree<Info, Tag> tr(t);
+    for (int i = 0; i < n; ++i) {
+        if (mark[i] == 0) {
+            tr.apply(get(a[i]), { a[i], 1 });
+        }
+    }
+
+    ll both_sum = 0;
+    for (int i = 0; i <= both.size(); ++i) {
+        if (i) both_sum += a[both[i - 1]];
+        int left_rem = max(0, k - i), right_rem = max(0, k - i);
+        while (left.size() > left_rem) {
+            int v = popback(left);
+            left_sum -= a[v];
+            tr.apply(get(a[v]), { a[v], 1 });
+        }
+        while (right.size() > right_rem) {
+            int v = popback(right);
+            right_sum -= a[v];
+            tr.apply(get(a[v]), { a[v], 1 });
+        }
+        if (left.size() < left_rem or right.size() < right_rem) {
+            continue;
+        }
+
+        int rem = m - (i + left_rem + right_rem);
+        if (rem < 0) continue;
+        int split = tr.binary_search(rem);
+        auto [sum, cnt] = tr.range_query(0, split);
+        if (cnt < rem) continue;
+        sum -= ll(cnt - rem) * oc[split];
+        chmin(res, both_sum + left_sum + right_sum + sum);
+    }
+
+    cout << (res == INFLL ? -1 : res) << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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-804d.cc

@@ -0,0 +1,784 @@
+// #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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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() {
+    constexpr int B = 316;
+    read(int, n, m, q);
+    adj(ch, n);
+    quick_union qu(n + 1);
+    for (int i = 0; i < m; ++i) {
+        read(int, u, v);
+        qu.merge(u, v);
+        edge(ch, u, v);
+    }
+
+    vector<pii> cc;
+    for (int i = 1; i <= n; ++i) {
+        if (qu.query(i) == i) {
+            cc.emplace_back(i, qu.query_size(i));
+        }
+    }
+
+    m = cc.size();
+
+    vector<int> diameter(n + 1);
+    {
+        int p, q;
+        int maxd;
+        auto dfs = [&] (auto dfs, int v, int pa, int d) -> void {
+            if (chmax(maxd, d)) {
+                q = v;
+            }
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                dfs(dfs, u, v, d + 1);
+            }
+        };
+        for (auto&& [root, cnt] : cc) {
+            p = root, maxd = -1;
+            dfs(dfs, p, 0, 0);
+            p = q, maxd = -1;
+            dfs(dfs, p, 0, 0);
+            diameter[root] = maxd;
+        }
+    }
+
+    vector<int> max_depth(n + 1);
+    vector<int> depth(n + 1);
+    vector<int> d(n + 1);
+    vector<int> idx(n + 1);
+    {
+        int j;
+        auto dfs = [&] (auto dfs, int v, int pa, int d) -> void {
+            idx[v] = j;
+            max_depth[v] = depth[v] = d;
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                dfs(dfs, u, v, d + 1);
+                chmax(max_depth[v], max_depth[u]);
+            }
+        };
+        for (int i = 0; i < m; ++i) {
+            auto&& [v, cnt] = cc[i];
+            j = i;
+            dfs(dfs, v, 0, 0);
+        }
+    }
+    {
+        auto dfs = [&] (auto dfs, int v, int pa, int up) -> void {
+            multiset<int> ds;
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                ds.emplace(max_depth[u] - depth[v]);
+            }
+            chmax(d[v], max(up, ds.empty() ? -INF : *ds.rbegin()));
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                ds.erase(ds.find(max_depth[u] - depth[v]));
+                dfs(dfs, u, v, max({up + 1, 1, ds.empty() ? -INF : *ds.rbegin() + 1}));
+                ds.emplace(max_depth[u] - depth[v]);
+            }
+        };
+        for (auto&& [v, cnt] : cc) {
+            dfs(dfs, v, 0, -INF);
+        }
+    }
+
+
+    auto work = [&] (int i, int j) -> ld {
+        auto&& [rootleft, totleft] = cc[i];
+        vector<int> cnt(totleft + 1);
+        {
+            auto dfs = [&] (auto dfs, int v, int pa) -> void {
+                cnt[d[v]] += 1;
+                for (auto&& u : ch[v]) {
+                    if (u == pa) continue;
+                    dfs(dfs, u, v);
+                }
+            };
+            dfs(dfs, rootleft, 0);
+        }
+        vector<int> ps(totleft + 2);
+        vector<ll> pssum(totleft + 2);
+        for (int i = 1; i <= totleft + 1; ++i) {
+            ps[i] = ps[i - 1] + cnt[i - 1];
+            pssum[i] = pssum[i - 1] + ll(1) * cnt[i - 1] * (i - 1);
+        }
+        auto&& [rootright, totright] = cc[j];
+        ld curr = 0;
+        int treed = max(diameter[rootleft], diameter[rootright]);
+        {
+            auto dfs = [&] (auto dfs, int v, int pa) -> void {
+                curr += ll(1) * ps[clamp(treed - d[v] - 1, 0, totleft + 1)] * treed + ll(1) * (ps[totleft + 1] - ps[clamp(treed - d[v] - 1, 0, totleft + 1)]) * (d[v] + 1) + (pssum[totleft + 1] - pssum[clamp(treed - d[v] - 1, 0, totleft + 1)]);
+                for (auto&& u : ch[v]) {
+                    if (u == pa) continue;
+                    dfs(dfs, u, v);
+                }
+            };
+            dfs(dfs, rootright, 0);
+        }
+        return ld(1) * curr / (ll(1) * totleft * totright);
+    };
+
+    vector<int> rev(m, -1);
+    int t = 0;
+    for (int i = 0; i < m; ++i) {
+        auto&& [v, tot] = cc[i];
+        if (tot < B) continue;
+        rev[i] = t++;
+    }
+    vector pre(m, vector<ld>(t));
+    // debug(t);
+    for (int i = 0; i < m; ++i) {
+        if (rev[i] == -1) continue;
+        auto&& [rootleft, totleft] = cc[i];
+        vector<int> cnt(n + 1);
+        {
+            auto dfs = [&] (auto dfs, int v, int pa) -> void {
+                cnt[d[v]] += 1;
+                for (auto&& u : ch[v]) {
+                    if (u == pa) continue;
+                    dfs(dfs, u, v);
+                }
+            };
+            dfs(dfs, rootleft, 0);
+        }
+        vector<int> ps(totleft + 2);
+        vector<ll> pssum(totleft + 2);
+        for (int i = 1; i <= totleft + 1; ++i) {
+            ps[i] = ps[i - 1] + cnt[i - 1];
+            pssum[i] = pssum[i - 1] + ll(1) * cnt[i - 1] * (i - 1);
+        }
+        for (int j = 0; j < m; ++j) {
+            if (j == i) continue;
+            auto&& [rootright, totright] = cc[j];
+            ld curr = 0;
+            int treed = max(diameter[rootleft], diameter[rootright]);
+            {
+                auto dfs = [&] (auto dfs, int v, int pa) -> void {
+                    curr += ll(1) * ps[clamp(treed - d[v] - 1, 0, totleft + 1)] * treed + ll(1) * (ps[totleft + 1] - ps[clamp(treed - d[v] - 1, 0, totleft + 1)]) * (d[v] + 1) + (pssum[totleft + 1] - pssum[clamp(treed - d[v] - 1, 0, totleft + 1)]);
+                    for (auto&& u : ch[v]) {
+                        if (u == pa) continue;
+                        dfs(dfs, u, v);
+                    }
+                };
+                dfs(dfs, rootright, 0);
+            }
+            pre[j][rev[i]] = ld(1) * curr / (ll(1) * totleft * totright);
+        }
+    }
+
+    cout << fixed << setprecision(50);
+    // int cnt = 0;
+    while (q--) {
+        read(int, u, v);
+        u = qu.query(u), v = qu.query(v);
+        if (u == v) {
+            cout << "-1\n";
+        } else if (max(cc[idx[u]].second, cc[idx[v]].second) < B) {
+            // deb(cc[idx[u]].second, cc[idx[v]].second);
+            // cnt += 1;
+            // debug(cnt);
+            cout << (long double)work(idx[u], idx[v]) << '\n';
+        } else {
+            if (cc[idx[v]].second >= B) {
+                swap(u, v);
+            }
+            cout << (long double)pre[idx[v]][rev[idx[u]]] << '\n';
+        }
+    }
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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/cp-templates

@@ -1 +1 @@
-Subproject commit 6ce8d08534507e063e06d58570725ec272166fd4
+Subproject commit b33b8ed3f4e64e62b354eb3acf0f92ec93a41f75

src/bin/d.cc

@@ -527,7 +527,7 @@ constexpr std::array<T, N> __initarray(const T& value) {
 }
 /*******************************************************/
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -545,7 +545,7 @@ template <typename T> struct point {
     point(const T& x, const T& y) : x(x), y(y) {}
 
     inline T square() const { return x * x + y * y; }
-    inline ld norm() const { return sqrt(double(square())); }
+    inline ld norm() const { return sqrt((long double)(square())); }
 
     inline point operator+(const point& rhs) const { return point(x + rhs.x, y + rhs.y); }
     inline point operator-(const point& rhs) const { return point(x - rhs.x, y - rhs.y); }
@@ -590,32 +590,60 @@ template <typename T> struct point {
 };
 
 
+struct convex_hull {
+    int n, m;
+    vector<int> on_hull;
+    // WARN: counter-clockwise
+    vector<int> hull;
+
+    // WARN: sort the points before initializing
+    template <typename T>
+    convex_hull(const vector<point<T>>& a) : n(a.size()), on_hull(n) {
+        for (int i = 0; i < n; ++i) {
+            int m = hull.size();
+            // WARN: keeping the points that are on edges
+            while (m > 1 and (a[hull[m - 1]] - a[hull[m - 2]]) * (a[i] - a[hull[m - 2]]) < 0) {
+                on_hull[hull[m - 1]] -= 1;
+                hull.pop_back();
+                m -= 1;
+            }
+            on_hull[i] += 1;
+            hull.emplace_back(i);
+        }
+        int low = hull.size();
+
+        for (int i = n - 2; ~i; --i) {
+            int m = hull.size();
+            while (m > low and (a[hull[m - 1]] - a[hull[m - 2]]) * (a[i] - a[hull[m - 2]]) < 0) {
+                on_hull[hull[m - 1]] -= 1;
+                hull.pop_back();
+                m -= 1;
+            }
+            on_hull[i] += 1;
+            hull.emplace_back(i);
+        }
+        if (n > 1) {
+            hull.pop_back();
+        }
+        m = hull.size();
+    }
+};
+
+
 // __attribute__((target("popcnt")))
 void solve() {
+    using point = point<ll>;
     read(int, n);
-    vector<pll> pts;
+    readvec(point, a, n);
-    for (int i = 0; i < n; ++i) {
+    vector<ld> dp(n + 1, -INFLL);
-        read(ll, x, y, u, v);
+    dp[0] = 0;
-        if (u - x == 0) {
+    for (int i = 1; i <= n; ++i) {
-            if (v > y) pts.emplace_back(0, 1);
+        for (int j = 1; j <= i; ++j) {
-            else pts.emplace_back(0, -1);
+            chmax(dp[i], dp[j - 1] + (a[i - 1] - a[j - 1]).norm());
-        } else if (v - y == 0) {
-            if (u < x) pts.emplace_back(-1, 0);
-            else pts.emplace_back(1, 0);
-        } else {
-            ll g = gcd(u - x, v - y);
-            pts.emplace_back((u - x) / g, (v - y) / g);
         }
     }
-    ll res= 0;
+    cout << fixed << setprecision(50);
-    unordered_map<pll, int, pair_hash> oc;
+    cout << (long double)dp[n] << '\n';
-    for (auto&& [x, y] : pts) {
-        if (oc.count({-x, -y})) {
-            res += oc[{-x, -y}];
-        }
-        oc[{x, y}] += 1;
-    }
-    cout << res << '\n';
 }
 
 int main() {

src/bin/e.cc

@@ -1,5 +1,4 @@
 // #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
-#include <algorithm>
 #pragma GCC optimize("Ofast,unroll-loops")
 /************* This code requires C++17. ***************/
 
@@ -528,7 +527,7 @@ constexpr std::array<T, N> __initarray(const T& value) {
 }
 /*******************************************************/
 
-#define SINGLE_TEST_CASE
+// #define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -541,15 +540,81 @@ void prep() {
 
 // __attribute__((target("popcnt")))
 void solve() {
-    read(int, n);
+    read(int, n, m);
-    vector<int> a(n);
+    if (n % 4 == 0 or m % 4 == 0) {
-    iota(a.begin(), a.end(), 1);
+        int s = 0;
-    int res = 0;
+        if (m % 4 == 0) swap(n, m), s = 1;
-    for (int i = 1; i <= n; ++i) res xor_eq i;
+        vector a(n, vector<int>(m));
-    cout << res << '\n';
+        int use = 0;
+        for (int i = 0; i < n; i += 4) {
+            for (int j = 0; j < m; ++j) {
+                use += 1;
+                for (int k = i; k < i + 4; ++k) {
+                    a[k][j] = use;
+                }
+            }
+        }
+        cout << "YES\n";
+        if (s) {
+            for (int i = 0; i < m; ++i) {
+                for (int j = 0; j < n; ++j) {
+                    cout << a[j][i] << ' ';
+                }
+                cout << '\n';
+            }
+        } else {
             for (int i = 0; i < n; ++i) {
-        cout << a << '\n';
+                for (int j = 0; j < m; ++j) {
-        rotate(a.begin(), a.begin() + 1, a.end());
+                    cout << a[i][j] << ' ';
+                }
+                cout << '\n';
+            }
+        }
+    } else if (n % 2 or m % 2 or max(n, m) == 2) {
+        cout << "NO\n";
+    } else {
+        int s = 0;
+        if (n > m) swap(n, m), s = 1;
+        vector a(n, vector<int>(m));
+        int use = 0;
+        for (int i = 0; i + 3 < n; i += 4) {
+            for (int j = 0; j < m; ++j) {
+                use += 1;
+                for (int k = i; k < i + 4; ++k) {
+                    a[k][j] = use;
+                }
+            }
+        }
+        for (int j = 0; j < m - 6; j += 4) {
+            for (int i = n - 2; i < n; ++i) {
+                use += 1;
+                for (int k = j; k < j + 4; ++k) {
+                    a[i][k] = use;
+                }
+            }
+        }
+        use += 1;
+        a[n - 2][m - 6] = a[n - 1][m - 6] = a[n - 1][m - 5] = a[n - 1][m - 4] = use;
+        use += 1;
+        a[n - 2][m - 5] = a[n - 2][m - 4] = a[n - 2][m - 3] = a[n - 2][m - 2] = use;
+        use += 1;
+        a[n - 2][m - 1] = a[n - 1][m - 3] = a[n - 1][m - 2] = a[n - 1][m - 1] = use;
+        cout << "YES\n";
+        if (s) {
+            for (int i = 0; i < m; ++i) {
+                for (int j = 0; j < n; ++j) {
+                    cout << a[j][i] << ' ';
+                }
+                cout << '\n';
+            }
+        } else {
+            for (int i = 0; i < n; ++i) {
+                for (int j = 0; j < m; ++j) {
+                    cout << a[i][j] << ' ';
+                }
+                cout << '\n';
+            }
+        }
     }
 }
 

src/bin/g.cc

@@ -1,5 +1,5 @@
 // #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
-// #pragma GCC optimize("Ofast,unroll-loops")
+#pragma GCC optimize("Ofast,unroll-loops")
 /************* This code requires C++17. ***************/
 
 #include<bits/stdc++.h>
@@ -527,7 +527,7 @@ constexpr std::array<T, N> __initarray(const T& value) {
 }
 /*******************************************************/
 
-#define SINGLE_TEST_CASE
+// #define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -538,160 +538,43 @@ 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 = 0;
-    void apply(const Tag& rhs, size_t len) { }
-};
-
-Info operator+(const Info &a, const Info &b) {
-    return {a.val + b.val};
-}
-
-
 // __attribute__((target("popcnt")))
 void solve() {
     read(int, n, k);
     readvec(int, a, n);
-    vector<int> del(n + 1, 1);
+    vector<pii> oc(n);
-    for (int i = 0; i < k; ++i) {
-        read(int, x);
-        del[x] = 0;
-    }
-    vector<int> rev(n + 1);
     for (int i = 0; i < n; ++i) {
-        rev[a[i]] = i;
+        oc[i] = {a[i], i};
     }
-    set<int> tr;
+    sort_by_key(oc.begin(), oc.end(), expr(pii(p.first, -p.second), auto&& p));
-    segtree<Info, Tag> cnt(n);
+    set<int> empty;
+    for (int i = 0; i < n; ++i) {
+        empty.emplace(i);
+    }
+    vector<int> stack;
+    int pre = -1;
     ll res = 0;
-    for (int i = 1; i <= n; ++i) {
+    for (auto&& [val, pos] : oc) {
-        if (del[i]) {
+        if (val != pre) stack = {};
-            int left = 0, right = n - 1;
+        pre = val;
-            auto it = tr.lower_bound(rev[i]);
+        stack.emplace_back(pos);
-            if (it != tr.end()) {
+        empty.erase(pos);
-                right = *it - 1;
+        int ll = 0, lr = pos, rl = pos, rr = n - 1;
+        {
+            auto it = empty.lower_bound(pos);
+            if (it != empty.begin()) {
+                ll = *prev(it) + 1;
             }
-            if (it != tr.begin()) {
-                left = *prev(it) + 1;
         }
-            // deb(cnt.range_query(left, right).val);
+        if (stack.size() >= k) {
-            // if (res > INT_MAX - right - left + 1 - cnt.range_query(left, right).val) {
+            rl = stack[stack.size() - k];
-            //     debug(res);
+            auto it = empty.lower_bound(lr);
-            // }
+            if (it != empty.end()) {
-            res += right - left + 1 - cnt.range_query(left, right).val;
+                rr = *it - 1;
-            cnt.set(rev[i], { 1 });
+            }
+            if (rl <= rr) {
+                res += 1LL * (lr - ll + 1) * (rr - rl + 1);
             }
-        if (not del[i]) {
-            tr.emplace(rev[i]);
         }
     }
     cout << res << '\n';

src/bin/gym-105481a.cc

@@ -0,0 +1,592 @@
+// #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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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() {
+    int n;
+    unordered_set<string> oc;
+    while (1) {
+        read(string, s);
+        if (s[0] >= '0' and s[0] <= '9') {
+            n = stoi(s);
+            break;
+        }
+        string nw;
+        for (auto&& c : s) {
+            if (c >= 'A' and c <= 'Z') {
+                nw += c - 'A' + 'a';
+            } else if (c >= 'a' and c <= 'z') {
+                nw += c;
+            }
+        }
+        oc.emplace(nw);
+    }
+    while (n--) {
+        read(string, s);
+        oc.erase(s);
+    }
+    cout << oc.size() << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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/gym-105481b.cc

@@ -0,0 +1,572 @@
+// #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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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(string, s);
+    string s1(s.rbegin(), s.rend());
+    cout << s1 << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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/gym-105481c.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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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);
+    vector<int> a(n + 1);
+    a[0] = 2200;
+    adj(ch, n);
+    for (int i = 1; i <= n; ++i) {
+        read(int, f, x);
+        edge(ch, i, f);
+        a[i] = x;
+    }
+    vector<ll> ps(n + 1);
+    vector<bool> valid(n + 1);
+    {
+        auto dfs = [&] (auto dfs, int v, int pa) -> void {
+            int son = 0;
+            for (auto&& u : ch[v]) {
+                if (u == pa) continue;
+                son += 1;
+                dfs(dfs, u, v);
+                ps[v] += ps[u];
+            }
+            if (not son) {
+                ps[v] = a[v];
+            } else {
+                valid[v] = 1;
+            }
+        };
+        dfs(dfs, 0, 0);
+    }
+    if (ps[0] > a[0]) {
+        cout << "NO\n";
+        return;
+    }
+
+    sort(a.begin() + 1, a.end());
+    sort(ps.begin() + 1, ps.end());
+    for (int i = 1; i <= n; ++i) {
+        if (ps[i] > a[i]) {
+            cout << "NO\n";
+            return;
+        }
+    }
+    cout << "YES\n";
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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/gym-105481d.cc

@@ -0,0 +1,673 @@
+// #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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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 T> struct point {
+    T x, y;
+    point() : x(), y() {}
+    point(const pair<T, T>& a) : x(a.first), y(a.second) {}
+    point(const T& x, const T& y) : x(x), y(y) {}
+
+    inline T square() const { return x * x + y * y; }
+    inline ld norm() const { return sqrt((long double)(square())); }
+
+    inline point operator+(const point& rhs) const { return point(x + rhs.x, y + rhs.y); }
+    inline point operator-(const point& rhs) const { return point(x - rhs.x, y - rhs.y); }
+    inline point operator+() const { return *this; }
+    inline point operator-() const { return point(-x, -y); }
+    inline point operator*(const T& a) const { return point(x * a, y * a); }
+    inline T operator*(const point& rhs) const { return x * rhs.y - y * rhs.x; }
+    inline point operator/(const T& a) const { return point(x / a, y / a); }
+    inline point& operator+=(const point& rhs) { x += rhs.x, y += rhs.y; return *this; }
+    inline point& operator-=(const point& rhs) { x -= rhs.x, y -= rhs.y; return *this; }
+    inline point& operator*=(const T& a) { x *= a, y *= a; return *this; }
+    inline point& operator/=(const T& a) { x /= a, y /= a; return *this; }
+
+    inline bool operator==(const point& rhs) const { return x == rhs.x and y == rhs.y; }
+    inline bool operator!=(const point& rhs) const { return not (*this == rhs); }
+    inline bool operator<(const point& rhs) const { return pair(x, y) < pair(rhs.x, rhs.y); }
+    inline bool operator<=(const point& rhs) const { return *this < rhs or *this == rhs; }
+    inline bool operator>(const point& rhs) const { return not (*this <= rhs); }
+    inline bool operator>=(const point& rhs) const { return not (*this < rhs); }
+
+    static inline ld slope(const point& a, const point& b) {
+        if (a.x == b.x) return INFLL;
+        return ld(a.y - b.y) / (a.x - b.x);
+    }
+
+    // distance from point `a` to line `l--r`
+    static inline ld dist(const point& a, const point& l, const point& r) {
+        return area(a, l, r) * 2 / (l - r).norm();
+    }
+
+    static inline ld area(const point& a, const point& b, const point& c) {
+        return (b - a) * (c - a) / ld(2);
+    }
+
+    friend inline istream& operator>>(istream& in, point& a) {
+        return in >> a.x >> a.y;
+    }
+
+    friend inline ostream& operator<<(ostream& out, const point& a) {
+        return out << a.x << ' ' << a.y;
+    }
+};
+
+
+struct convex_hull {
+    int n, m;
+    vector<int> on_hull;
+    // WARN: counter-clockwise
+    vector<int> hull;
+
+    // WARN: sort the points before initializing
+    template <typename T>
+    convex_hull(const vector<point<T>>& a) : n(a.size()), on_hull(n) {
+        for (int i = 0; i < n; ++i) {
+            int m = hull.size();
+            // WARN: keeping the points that are on edges
+            while (m > 1 and (a[hull[m - 1]] - a[hull[m - 2]]) * (a[i] - a[hull[m - 2]]) < 0) {
+                on_hull[hull[m - 1]] -= 1;
+                hull.pop_back();
+                m -= 1;
+            }
+            on_hull[i] += 1;
+            hull.emplace_back(i);
+        }
+        int low = hull.size();
+
+        for (int i = n - 2; ~i; --i) {
+            int m = hull.size();
+            while (m > low and (a[hull[m - 1]] - a[hull[m - 2]]) * (a[i] - a[hull[m - 2]]) < 0) {
+                on_hull[hull[m - 1]] -= 1;
+                hull.pop_back();
+                m -= 1;
+            }
+            on_hull[i] += 1;
+            hull.emplace_back(i);
+        }
+        if (n > 1) {
+            hull.pop_back();
+        }
+        m = hull.size();
+    }
+};
+
+
+// __attribute__((target("popcnt")))
+void solve() {
+    using point = point<ll>;
+    read(int, n);
+    readvec(point, a, n);
+    vector<ld> dp(n + 1, -INFLL);
+    dp[0] = 0;
+    for (int i = 1; i <= n; ++i) {
+        for (int j = 1; j <= i; ++j) {
+            chmax(dp[i], dp[j - 1] + (a[i - 1] - a[j - 1]).norm());
+        }
+    }
+    cout << fixed << setprecision(50);
+    cout << (long double)dp[n] << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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/gym-105481e.cc

@@ -0,0 +1,645 @@
+// #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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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, m);
+    if (n % 4 == 0 or m % 4 == 0) {
+        int s = 0;
+        if (m % 4 == 0) swap(n, m), s = 1;
+        vector a(n, vector<int>(m));
+        int use = 0;
+        for (int i = 0; i < n; i += 4) {
+            for (int j = 0; j < m; ++j) {
+                use += 1;
+                for (int k = i; k < i + 4; ++k) {
+                    a[k][j] = use;
+                }
+            }
+        }
+        cout << "YES\n";
+        if (s) {
+            for (int i = 0; i < m; ++i) {
+                for (int j = 0; j < n; ++j) {
+                    cout << a[j][i] << ' ';
+                }
+                cout << '\n';
+            }
+        } else {
+            for (int i = 0; i < n; ++i) {
+                for (int j = 0; j < m; ++j) {
+                    cout << a[i][j] << ' ';
+                }
+                cout << '\n';
+            }
+        }
+    } else if (n % 2 or m % 2 or max(n, m) == 2) {
+        cout << "NO\n";
+    } else {
+        int s = 0;
+        if (n > m) swap(n, m), s = 1;
+        vector a(n, vector<int>(m));
+        int use = 0;
+        for (int i = 0; i + 3 < n; i += 4) {
+            for (int j = 0; j < m; ++j) {
+                use += 1;
+                for (int k = i; k < i + 4; ++k) {
+                    a[k][j] = use;
+                }
+            }
+        }
+        for (int j = 0; j < m - 6; j += 4) {
+            for (int i = n - 2; i < n; ++i) {
+                use += 1;
+                for (int k = j; k < j + 4; ++k) {
+                    a[i][k] = use;
+                }
+            }
+        }
+        use += 1;
+        a[n - 2][m - 6] = a[n - 1][m - 6] = a[n - 1][m - 5] = a[n - 1][m - 4] = use;
+        use += 1;
+        a[n - 2][m - 5] = a[n - 2][m - 4] = a[n - 2][m - 3] = a[n - 2][m - 2] = use;
+        use += 1;
+        a[n - 2][m - 1] = a[n - 1][m - 3] = a[n - 1][m - 2] = a[n - 1][m - 1] = use;
+        cout << "YES\n";
+        if (s) {
+            for (int i = 0; i < m; ++i) {
+                for (int j = 0; j < n; ++j) {
+                    cout << a[j][i] << ' ';
+                }
+                cout << '\n';
+            }
+        } else {
+            for (int i = 0; i < n; ++i) {
+                for (int j = 0; j < m; ++j) {
+                    cout << a[i][j] << ' ';
+                }
+                cout << '\n';
+            }
+        }
+    }
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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/gym-105481g.cc

@@ -0,0 +1,607 @@
+// #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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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, k);
+    readvec(int, a, n);
+    vector<pii> oc(n);
+    for (int i = 0; i < n; ++i) {
+        oc[i] = {a[i], i};
+    }
+    sort_by_key(oc.begin(), oc.end(), expr(pii(p.first, -p.second), auto&& p));
+    set<int> empty;
+    for (int i = 0; i < n; ++i) {
+        empty.emplace(i);
+    }
+    vector<int> stack;
+    int pre = -1;
+    ll res = 0;
+    for (auto&& [val, pos] : oc) {
+        if (val != pre) stack = {};
+        pre = val;
+        stack.emplace_back(pos);
+        empty.erase(pos);
+        int ll = 0, lr = pos, rl = pos, rr = n - 1;
+        {
+            auto it = empty.lower_bound(pos);
+            if (it != empty.begin()) {
+                ll = *prev(it) + 1;
+            }
+        }
+        if (stack.size() >= k) {
+            rl = stack[stack.size() - k];
+            auto it = empty.lower_bound(lr);
+            if (it != empty.end()) {
+                rr = *it - 1;
+            }
+            if (rl <= rr) {
+                res += 1LL * (lr - ll + 1) * (rr - rl + 1);
+            }
+        }
+    }
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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/gym-105481j.cc

@@ -0,0 +1,579 @@
+// #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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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);
+    read(int, a, b, c);
+    int res = 0;
+    readvec(pii, s, n);
+    read(int, d);
+    for (auto&& [x, y] : s) {
+        if (x + y >= c) continue;
+        if (min(a, x + d) + y >= c) res += 1;
+    }
+    cout << res<< '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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/gym-105481k.cc

@@ -0,0 +1,624 @@
+// #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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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() {
+}
+
+constexpr int N = 5e5 + 1;
+constexpr int M = 5e3 + 1;
+vector<int> d[M << 2];
+bitset<N> blob[M];
+vector<int> oc[N];
+
+// __attribute__((target("popcnt")))
+void solve() {
+    read(int, n);
+    auto range_apply = [&] (auto range_apply, int s, int t, int p, int l, int r, int x) -> void {
+        if (l <= s and t <= r) {
+            d[p].emplace_back(x);
+            return;
+        }
+        int m = s + (t - s >> 1);
+        if (l <= m) range_apply(range_apply, s, m, p * 2, l, r, x);
+        if (r > m) range_apply(range_apply, m + 1, t, p * 2 + 1, l, r, x);
+    };
+    int tot = 1;
+    vector<int> idx(n << 2, -1);
+    idx[1] = 0;
+    vector<int> res(n);
+    deque<tiii> q;
+    readvec(pii, a, n);
+    for (int i = 0; i < n; ++i) {
+        auto [op, x] = a[i];
+        if (op == 1) {
+            oc[x].emplace_back(i);
+        } else {
+            range_apply(range_apply, 0, n - 1, 1, popback(oc[x]), i - 1, x);
+        }
+    }
+    for (int i = 0; i < N; ++i) {
+        while (oc[i].size()) {
+            range_apply(range_apply, 0, n - 1, 1, popback(oc[i]), n - 1, i);
+        }
+    }
+    q.emplace_back(0, n - 1, 1);
+    while (q.size()) {
+        auto [s, t, p] = popfront(q);
+        blob[idx[p]][0] = 1;
+        for (auto&& x : d[p]) {
+            blob[idx[p]] |= blob[idx[p]] << x;
+        }
+        if (s == t) {
+            res[s] = blob[idx[p]].count() - 1;
+            continue;
+        }
+        int m = s + (t - s >> 1);
+        idx[p * 2] = idx[p];
+        blob[tot++] = blob[idx[p]];
+        idx[p * 2 + 1] = tot - 1;
+        q.emplace_back(s, m, p * 2);
+        q.emplace_back(m + 1, t, p * 2 + 1);
+    }
+    // debug(tot);
+    putvec_eol(res);
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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/gym-105481l.cc

@@ -0,0 +1,596 @@
+// #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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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(ll, k);
+
+    auto work = [&] (int128 tot) -> int128 {
+        constexpr int N = 10;
+        int128 res = 0;
+        int128 pw = 1;
+        for (int i = 1; i <= N; ++i) {
+            pw *= 100;
+        }
+        for (int i = N; ~i; --i) {
+            res += tot / (4 * pw) - tot / (100 * pw);
+            pw /= 100;
+        }
+        return res;
+    };
+
+    int128 l = 2024, r = LLONG_MAX;
+    int128 start = 2024 - work(2024);
+    while (l < r) {
+        int128 mid = l + r >> 1;
+        if ((mid - work(mid)) - start >= k) {
+            r = mid;
+        } else {
+            l = mid + 1;
+        }
+    }
+    cout << l << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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/gym-105481m.cc

@@ -0,0 +1,801 @@
+// #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 = __float128;  // up to 1e-9 precision in binary search
+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 PRIMELL = 901017227882342239LL;
+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 {
+    safe_hash hasher;
+    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) {
+            auto h = hasher(x);
+            res1 = (res1 + h * pw1) % __array_hash_mdl1;
+            res2 = (res2 + h * 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 circular 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, (long double)(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(const string& s, const 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(const 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(int128(lhs.val) * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return lhs * mod(inverse(rhs.val, 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, m, k, t);
+    readvec(int, a, k);
+    vector<int> cnt(m + 1);
+    int l = 0, r = n;
+    vector grid(3, vector<int>(3, -1));
+    int extra = 0;
+
+    vector<vector<int>> mp = {{2, 3, 4}, {5, 1, 6}, {7, 8, 9}};
+    vector<pii> seq = {{1, 1}, {0, 0}, {0, 1}, {0, 2}, {1, 0}, {1, 2}, {2, 0}, {2, 1}, {2, 2}};
+    while (1) {
+        int f = 1;
+        // debug(0);
+        // debug(grid);
+
+        int op = 0;
+
+        for (auto&& [i, j] : seq) {
+            // deb(i, j, l, r);
+            if (grid[i][j] == -1 and l < r) {
+                grid[i][j] = a[l++];
+                op = 1;
+                if (grid[i][j] == t) {
+                    if (r == k) {
+                        extra += 1;
+                    } else {
+                        r += 1;
+                    }
+                } else if (grid[i][j] == 0) {
+                    cnt[grid[i][j]] += 1;
+                    grid[i][j] = -1;
+                    if (r == k) {
+                        extra += 1;
+                    } else {
+                        r += 1;
+                    }
+                    f = 0;
+                    goto fi;
+                }
+            }
+        }
+        // debug(2);
+        // debug(grid);
+fi:
+        if (f == 0) continue;
+        unordered_set<int> oc;
+        for (int i = 0; i < 3; ++i) {
+            for (int j = 0; j < 3; ++j) {
+                if (grid[i][j] == -1 or oc.count(grid[i][j])) {
+                    f = 0;
+                    goto fii;
+                }
+                oc.emplace(grid[i][j]);
+            }
+        }
+fii:
+        if (f == 1) {
+            cnt[grid[1][1]] += 1;
+            grid[1][1] = -1;
+            continue;
+        }
+        int use4 = 0;
+
+        // debug(1);
+        // debug(grid);
+        while (1) {
+            tiii cand = { INF, INF, INF };
+            vector mx = { INF, INF, INF };
+            for (int i = 0; i < 3; ++i) {
+                int f = 1;
+                for (int j = 0; j < 3; ++j) {
+                    if (grid[i][j] == -1) {
+                        f = 0;
+                        break;
+                    } else if (grid[i][j] != grid[i][0]) {
+                        f = 0;
+                        break;
+                    }
+                }
+                if (f) {
+                    vector curr = {mp[i][0], mp[i][1], mp[i][2]};
+                    sort(curr.begin(), curr.end());
+                    if (chmin(mx, curr)) {
+                        cand = tiii {i * 3, i * 3 + 1, i * 3 + 2};
+                    }
+                }
+            }
+            for (int j = 0; j < 3; ++j) {
+                int f = 1;
+                for (int i = 0; i < 3; ++i) {
+                    if (grid[i][j] == -1) {
+                        f = 0;
+                        break;
+                    } else if (grid[i][j] != grid[0][j]) {
+                        f = 0;
+                        break;
+                    }
+                }
+                if (f) {
+                    vector curr = { mp[0][j], mp[1][j], mp[2][j] };
+                    sort(curr.begin(), curr.end());
+                    if (chmin(mx, curr)) {
+                        cand = tiii {j, 3 + j, 6 + j};
+                    }
+                }
+            }
+            if (grid[0][0] == grid[1][1] and grid[0][0] == grid[2][2] and grid[0][0] != -1) {
+                vector curr = {mp[0][0], mp[1][1], mp[2][2]};
+                sort(curr.begin(), curr.end());
+                if (chmin(mx, curr)) {
+                    cand = tiii{0, 4, 8};
+                }
+            }
+            if (grid[0][2] == grid[1][1] and grid[0][2] == grid[2][0] and grid[0][2] != -1) {
+                vector curr = {mp[0][2], mp[1][1], mp[0][2]};
+                sort(curr.begin(), curr.end());
+                if (chmin(mx, curr)) {
+                    cand = tiii{2, 4, 6};
+                }
+            }
+            if (get<0>(cand) == INF) break;
+            op = 1;
+            for (int i = 0; i < 5; ++i) {
+                if (r == k) {
+                    extra += 1;
+                } else {
+                    r += 1;
+                }
+            }
+            auto [i, j, k] = cand;
+            // deb(i, j, k);
+            if (i == 4) {
+                use4 = 1;
+            } else {
+                cnt[grid[i / 3][i % 3]] += 1;
+                grid[i / 3][i % 3] = -1;
+            }
+            if (j == 4) {
+                use4 = 1;
+            } else {
+                cnt[grid[j / 3][j % 3]] += 1;
+                grid[j / 3][j % 3] = -1;
+            }
+            if (k == 4) {
+                use4 = 1;
+            } else {
+                cnt[grid[k / 3][k % 3]] += 1;
+                grid[k / 3][k % 3] = -1;
+            }
+        }
+
+        while (1) {
+            pii cand = { INF, INF };
+            vector mx = { INF, INF };
+            for (int i = 0; i < 3; ++i) {
+                for (int j = 0; j < 3; ++j) {
+                    if (grid[i][j] == -1) continue;
+                    for (int k = 0; k < 3; ++k) {
+                        for (int l = 0; l < 3; ++l) {
+                            if (k == i and l == j) continue;
+                            if (grid[k][l] == grid[i][j]) {
+                                vector curr = {mp[k][l], mp[i][j]};
+                                sort(curr.begin(), curr.end());
+                                if (chmin(mx, curr)) {
+                                    cand = pii{ i * 3 + j, k * 3 + l };
+                                }
+                            }
+                        }
+                    }
+                }
+            }
+            if (cand.first == INF) break;
+            auto [i, j] = cand;
+            op = 1;
+            for (int i = 0; i < 1; ++i) {
+                if (r == k) {
+                    extra += 1;
+                } else {
+                    r += 1;
+                }
+            }
+            if (i == 4) {
+                use4 = 1;
+            } else {
+                cnt[grid[i / 3][i % 3]] += 1;
+                grid[i / 3][i % 3] = -1;
+            }
+            if (j == 4) {
+                use4 = 1;
+            } else {
+                cnt[grid[j / 3][j % 3]] += 1;
+                grid[j / 3][j % 3] = -1;
+            }
+        }
+
+        if (use4) {
+            cnt[grid[1][1]] += 1;
+            grid[1][1] = -1;
+        }
+
+        if (op == 0) break;
+
+        int has = 0;
+        for (int i = 0; i < 3; ++i) {
+            for (int j = 0; j < 3; ++j) {
+                if (grid[i][j] != -1) {
+                    has = 1;
+                }
+            }
+        }
+        if (has == 0) {
+            for (int i = 0; i < 10; ++i) {
+                if (r == k) {
+                    extra += 1;
+                } else {
+                    r += 1;
+                }
+            }
+        }
+    }
+
+    for (int i = 0; i < 3; ++i) {
+        for (int j = 0; j < 3; ++j) {
+            if (grid[i][j] != -1) {
+                cnt[grid[i][j]] += 1;
+            }
+        }
+    }
+
+    putvec(cnt);
+    if (extra > 0) {
+        cout << "Unhappy! " << extra << endl;
+    }
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    static_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/j.cc

@@ -527,7 +527,7 @@ constexpr std::array<T, N> __initarray(const T& value) {
 }
 /*******************************************************/
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -535,45 +535,21 @@ void dump() {}
 
 void dump_ignore() {}
 
-constexpr int N = 2e5 + 10;
-ll fact[N], factrev[N + 1], s[N + 1];
-
 void prep() {
-    fact[0] = factrev[0] = 1;
-    for (int i = 1; i < N; ++i) {
-        fact[i] = (fact[i - 1] * i) % MDL;
-    }
-    s[0] = 1;
-    for (int i = 1; i <= N; ++i) {
-        s[i] = s[i - 1] * fact[i - 1] % MDL;
-    }
-    factrev[N] = inverse(s[N], MDL);
-    for (int i = N; i; --i) {
-        factrev[i - 1] = factrev[i] * fact[i - 1] % MDL;
-    }
-    for (int i = 0; i < N; ++i) {
-        factrev[i] = factrev[i + 1] * s[i] % MDL;
-    }
 }
 
 // __attribute__((target("popcnt")))
 void solve() {
-    read(int, n, p, q);
+    read(int, n);
-    ll win = p * inverse(q, MDL) % MDL;
+    read(int, a, b, c);
-    ll lose = mod(1 - win, MDL);
+    int res = 0;
-    ll res = 0;
+    readvec(pii, s, n);
-    ll fixed = qpow(win, n, MDL);
+    read(int, d);
-    auto mycomb = [&] (int n, int m) {
+    for (auto&& [x, y] : s) {
-        if (n < m or m < 0 or n < 0) return 0LL;
+        if (x + y >= c) continue;
-        return (fact[n] * factrev[m] % MDL) * factrev[n - m] % MDL;
+        if (min(a, x + d) + y >= c) res += 1;
-    };
-    ll pw = 1;
-    for (int i = 0; i < n; ++i) {
-        res += (mycomb(n - 1 + i, n - 1) * fixed % MDL) * pw % MDL;
-        res %= MDL;
-        pw = pw * lose % MDL;
     }
-    cout << res << '\n';
+    cout << res<< '\n';
 }
 
 int main() {

src/bin/k.cc

@@ -527,7 +527,7 @@ constexpr std::array<T, N> __initarray(const T& value) {
 }
 /*******************************************************/
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -538,30 +538,63 @@ void dump_ignore() {}
 void prep() {
 }
 
+constexpr int N = 5e5 + 1;
+constexpr int M = 5e3 + 1;
+vector<int> d[M << 2];
+bitset<N> blob[M];
+vector<int> oc[N];
+
 // __attribute__((target("popcnt")))
 void solve() {
-    using mll = MLL<MDL>;
+    read(int, n);
-    read(int, n, m);
+    auto range_apply = [&] (auto range_apply, int s, int t, int p, int l, int r, int x) -> void {
-    vector<int> diff(n + 1);
+        if (l <= s and t <= r) {
-    vector<int> diff2(n + 1);
+            d[p].emplace_back(x);
-    while (m--) {
+            return;
-        read(int, op, x);
+        }
+        int m = s + (t - s >> 1);
+        if (l <= m) range_apply(range_apply, s, m, p * 2, l, r, x);
+        if (r > m) range_apply(range_apply, m + 1, t, p * 2 + 1, l, r, x);
+    };
+    int tot = 1;
+    vector<int> idx(n << 2, -1);
+    idx[1] = 0;
+    vector<int> res(n);
+    deque<tiii> q;
+    readvec(pii, a, n);
+    for (int i = 0; i < n; ++i) {
+        auto [op, x] = a[i];
         if (op == 1) {
-            diff[x] += 1;
+            oc[x].emplace_back(i);
         } else {
-            diff2[x] += 1;
+            range_apply(range_apply, 0, n - 1, 1, popback(oc[x]), i - 1, x);
         }
     }
-    mll res = 0;
+    for (int i = 0; i < N; ++i) {
-    mll cnt = 0;
+        while (oc[i].size()) {
-    mll cnt2 = 0;
+            range_apply(range_apply, 0, n - 1, 1, popback(oc[i]), n - 1, i);
-    mll sum2 = 0;
-    for (int i = 1; i <= n; ++i) {
-        cnt += diff[i], cnt2 += diff2[i];
-        sum2 += cnt2;
-        res += cnt + (2 * sum2 - cnt2);
-        cout << res << " \n"[i == n];
         }
+    }
+    q.emplace_back(0, n - 1, 1);
+    while (q.size()) {
+        auto [s, t, p] = popfront(q);
+        blob[idx[p]][0] = 1;
+        for (auto&& x : d[p]) {
+            blob[idx[p]] |= blob[idx[p]] << x;
+        }
+        if (s == t) {
+            res[s] = blob[idx[p]].count() - 1;
+            continue;
+        }
+        int m = s + (t - s >> 1);
+        idx[p * 2] = idx[p];
+        blob[tot++] = blob[idx[p]];
+        idx[p * 2 + 1] = tot - 1;
+        q.emplace_back(s, m, p * 2);
+        q.emplace_back(m + 1, t, p * 2 + 1);
+    }
+    // debug(tot);
+    putvec_eol(res);
 }
 
 int main() {

src/bin/l.cc

@@ -527,7 +527,7 @@ constexpr std::array<T, N> __initarray(const T& value) {
 }
 /*******************************************************/
 
-#define SINGLE_TEST_CASE
+// #define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -540,64 +540,33 @@ void prep() {
 
 // __attribute__((target("popcnt")))
 void solve() {
-    read(int, n, m);
+    read(ll, k);
-    readvec(int, a, n);
-    readvec(int, b, m);
 
-    vector res(n, vector<int>(m));
+    auto work = [&] (int128 tot) -> int128 {
-    for (int i = 0; i < n; ++i) {
+        constexpr int N = 10;
-        for (int j = 0; j < m; ++j) {
+        int128 res = 0;
-            res[i][j] = b[j];
+        int128 pw = 1;
-        }
+        for (int i = 1; i <= N; ++i) {
-    }
+            pw *= 100;
-    int g = 0;
-    for (int i = 0; i < n; ++i) {
-        g = g == 0 ? a[i] : gcd(g, a[i]);
-    }
-    ll l = 1;
-    for (int j = 0; j < m; ++j) {
-        l = lcm(l, ll(1) * b[j]);
-        if (l > 1e9) {
-            cout << "No\n";
-            return;
-        }
-    }
-    if (g % l != 0) {
-        cout << "No\n";
-        return;
-    }
-    int col = 0, dir = 1;
-    for (int i = 0; i < n; ++i) {
-        if (l == a[i]) continue;
-        res[i][col] = a[i];
-        col += dir;
-        if (col == m) {
-            col = max(0, m - 2);
-            dir = -dir;
-        } else if (col == -1) {
-            col = min(m - 1, 1);
-            dir = -dir;
         }
+        for (int i = N; ~i; --i) {
+            res += tot / (4 * pw) - tot / (100 * pw);
+            pw /= 100;
         }
+        return res;
+    };
 
-    for (int j = 0; j < m; ++j) {
+    int128 l = 2024, r = LLONG_MAX;
-        int g = 0;
+    int128 start = 2024 - work(2024);
-        for (int i = 0; i < n; ++i) {
+    while (l < r) {
-            g = g == 0 ? res[i][j] : gcd(g, res[i][j]);
+        int128 mid = l + r >> 1;
-        }
+        if ((mid - work(mid)) - start >= k) {
-        if (g != b[j]) {
+            r = mid;
-            cout << "No\n";
+        } else {
-            return;
+            l = mid + 1;
         }
     }
-
+    cout << l << '\n';
-    cout << "Yes\n";
-    for (int i = 0; i < n; ++i) {
-        for (int j = 0; j < m; ++j) {
-            cout << res[i][j] << ' ';
-        }
-        cout << '\n';
-    }
 }
 
 int main() {

src/bin/m.cc

@@ -1,5 +1,5 @@
 // #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
-#pragma GCC optimize("O2,unroll-loops")
+#pragma GCC optimize("Ofast,unroll-loops")
 /************* This code requires C++17. ***************/
 
 #include<bits/stdc++.h>
@@ -527,7 +527,7 @@ constexpr std::array<T, N> __initarray(const T& value) {
 }
 /*******************************************************/
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -535,42 +535,243 @@ void dump() {}
 
 void dump_ignore() {}
 
-constexpr int B = 2;
-constexpr int N = 1e5;
-constexpr int s = 31623;  // sqrt(MDL)
-int a[N], b[N];
-
 void prep() {
-    a[0] = b[0] = 1;
-    for(int i = 1; i <= s; i++) a[i] = (ll)a[i - 1] * B % MDL ;
-    for(int i = 1; i <= s; i++) b[i] = (ll)b[i - 1] * a[s] % MDL ;
-}
-
-ll lpow(ll power) { return ll(b[power / s]) * a[power % s] % MDL; }
-// ll lpow(ll power) {
-//     return qpow(2, power, MDL);
-// }
-
-inline ll series(int n) {
-    return mod(1 - inverse(lpow(n + 1), MDL), MDL) * inverse(1 - inverse(2, MDL), MDL) % MDL;
 }
 
 // __attribute__((target("popcnt")))
 void solve() {
-    read(int, n);
+    read(int, n, m, k, t);
+    readvec(int, a, k);
+    vector<int> cnt(m + 1);
+    int l = 0, r = n;
+    vector grid(3, vector<int>(3, -1));
+    int extra = 0;
 
-    ll inv2 = inverse(2, MDL);
+    vector<vector<int>> mp = {{2, 3, 4}, {5, 1, 6}, {7, 8, 9}};
+    vector<pii> seq = {{1, 1}, {0, 0}, {0, 1}, {0, 2}, {1, 0}, {1, 2}, {2, 0}, {2, 1}, {2, 2}};
+    while (1) {
+        int f = 1;
+        // debug(0);
+        // debug(grid);
 
-    auto calc = [&] (int x) {
+        int op = 0;
-        ll left = ((n - 1) * lpow(n - 2) % MDL) * series(x) % MDL;
+
-        ll right = lpow(n - 2) * mod(-x * inverse(lpow(x), MDL) % MDL + series(x - 1), MDL) % MDL;
+        for (auto&& [i, j] : seq) {
-        ll mid = lpow(n) * series(x) % MDL;
+            // deb(i, j, l, r);
-        // mll mid = 0;
+            if (grid[i][j] == -1 and l < r) {
-        // deb(x, mid);
+                grid[i][j] = a[l++];
-        return mod(((left - right) % MDL + mid) % MDL - 2 * x, MDL);
+                op = 1;
-    };
+                if (grid[i][j] == t) {
-    // deb(calc(1), calc(3));
+                    if (r == k) {
-    cout << mod(mod(lpow(n - 1) - 1 - (n - 1), MDL) - mod(calc(n - 1) - calc(max(1, (n + 1) / 2 - 1)), MDL), MDL) << '\n';
+                        extra += 1;
+                    } else {
+                        r += 1;
+                    }
+                } else if (grid[i][j] == 0) {
+                    cnt[grid[i][j]] += 1;
+                    grid[i][j] = -1;
+                    if (r == k) {
+                        extra += 1;
+                    } else {
+                        r += 1;
+                    }
+                    f = 0;
+                    goto fi;
+                }
+            }
+        }
+        // debug(2);
+        // debug(grid);
+fi:
+        if (f == 0) continue;
+        unordered_set<int> oc;
+        for (int i = 0; i < 3; ++i) {
+            for (int j = 0; j < 3; ++j) {
+                if (grid[i][j] == -1 or oc.count(grid[i][j])) {
+                    f = 0;
+                    goto fii;
+                }
+                oc.emplace(grid[i][j]);
+            }
+        }
+fii:
+        if (f == 1) {
+            cnt[grid[1][1]] += 1;
+            grid[1][1] = -1;
+            continue;
+        }
+        int use4 = 0;
+
+        // debug(1);
+        // debug(grid);
+        while (1) {
+            tiii cand = { INF, INF, INF };
+            vector mx = { INF, INF, INF };
+            for (int i = 0; i < 3; ++i) {
+                int f = 1;
+                for (int j = 0; j < 3; ++j) {
+                    if (grid[i][j] == -1) {
+                        f = 0;
+                        break;
+                    } else if (grid[i][j] != grid[i][0]) {
+                        f = 0;
+                        break;
+                    }
+                }
+                if (f) {
+                    vector curr = {mp[i][0], mp[i][1], mp[i][2]};
+                    sort(curr.begin(), curr.end());
+                    if (chmin(mx, curr)) {
+                        cand = tiii {i * 3, i * 3 + 1, i * 3 + 2};
+                    }
+                }
+            }
+            for (int j = 0; j < 3; ++j) {
+                int f = 1;
+                for (int i = 0; i < 3; ++i) {
+                    if (grid[i][j] == -1) {
+                        f = 0;
+                        break;
+                    } else if (grid[i][j] != grid[0][j]) {
+                        f = 0;
+                        break;
+                    }
+                }
+                if (f) {
+                    vector curr = { mp[0][j], mp[1][j], mp[2][j] };
+                    sort(curr.begin(), curr.end());
+                    if (chmin(mx, curr)) {
+                        cand = tiii {j, 3 + j, 6 + j};
+                    }
+                }
+            }
+            if (grid[0][0] == grid[1][1] and grid[0][0] == grid[2][2] and grid[0][0] != -1) {
+                vector curr = {mp[0][0], mp[1][1], mp[2][2]};
+                sort(curr.begin(), curr.end());
+                if (chmin(mx, curr)) {
+                    cand = tiii{0, 4, 8};
+                }
+            }
+            if (grid[0][2] == grid[1][1] and grid[0][2] == grid[2][0] and grid[0][2] != -1) {
+                vector curr = {mp[0][2], mp[1][1], mp[0][2]};
+                sort(curr.begin(), curr.end());
+                if (chmin(mx, curr)) {
+                    cand = tiii{2, 4, 6};
+                }
+            }
+            if (get<0>(cand) == INF) break;
+            op = 1;
+            for (int i = 0; i < 5; ++i) {
+                if (r == k) {
+                    extra += 1;
+                } else {
+                    r += 1;
+                }
+            }
+            auto [i, j, k] = cand;
+            // deb(i, j, k);
+            if (i == 4) {
+                use4 = 1;
+            } else {
+                cnt[grid[i / 3][i % 3]] += 1;
+                grid[i / 3][i % 3] = -1;
+            }
+            if (j == 4) {
+                use4 = 1;
+            } else {
+                cnt[grid[j / 3][j % 3]] += 1;
+                grid[j / 3][j % 3] = -1;
+            }
+            if (k == 4) {
+                use4 = 1;
+            } else {
+                cnt[grid[k / 3][k % 3]] += 1;
+                grid[k / 3][k % 3] = -1;
+            }
+        }
+
+        while (1) {
+            pii cand = { INF, INF };
+            vector mx = { INF, INF };
+            for (int i = 0; i < 3; ++i) {
+                for (int j = 0; j < 3; ++j) {
+                    if (grid[i][j] == -1) continue;
+                    for (int k = 0; k < 3; ++k) {
+                        for (int l = 0; l < 3; ++l) {
+                            if (k == i and l == j) continue;
+                            if (grid[k][l] == grid[i][j]) {
+                                vector curr = {mp[k][l], mp[i][j]};
+                                sort(curr.begin(), curr.end());
+                                if (chmin(mx, curr)) {
+                                    cand = pii{ i * 3 + j, k * 3 + l };
+                                }
+                            }
+                        }
+                    }
+                }
+            }
+            if (cand.first == INF) break;
+            auto [i, j] = cand;
+            op = 1;
+            for (int i = 0; i < 1; ++i) {
+                if (r == k) {
+                    extra += 1;
+                } else {
+                    r += 1;
+                }
+            }
+            if (i == 4) {
+                use4 = 1;
+            } else {
+                cnt[grid[i / 3][i % 3]] += 1;
+                grid[i / 3][i % 3] = -1;
+            }
+            if (j == 4) {
+                use4 = 1;
+            } else {
+                cnt[grid[j / 3][j % 3]] += 1;
+                grid[j / 3][j % 3] = -1;
+            }
+        }
+
+        if (use4) {
+            cnt[grid[1][1]] += 1;
+            grid[1][1] = -1;
+        }
+
+        if (op == 0) break;
+
+        int has = 0;
+        for (int i = 0; i < 3; ++i) {
+            for (int j = 0; j < 3; ++j) {
+                if (grid[i][j] != -1) {
+                    has = 1;
+                }
+            }
+        }
+        if (has == 0) {
+            for (int i = 0; i < 10; ++i) {
+                if (r == k) {
+                    extra += 1;
+                } else {
+                    r += 1;
+                }
+            }
+        }
+    }
+
+    for (int i = 0; i < 3; ++i) {
+        for (int j = 0; j < 3; ++j) {
+            if (grid[i][j] != -1) {
+                cnt[grid[i][j]] += 1;
+            }
+        }
+    }
+
+    putvec(cnt);
+    if (extra > 0) {
+        cout << "Unhappy! " << extra << endl;
+    }
 }
 
 int main() {

src/bin/test.py

@@ -8,7 +8,10 @@ import io
 PRIME = 998_244_353
 
 if __name__ == '__main__':
-    N = 1000000
+    N = 100000
-    print(N, 0)
+    M = N // 2
-    print(*range(1, N + 1))
+    print(N, M, N)
-
+    for i in range(1, M + 1):
+        print(randint(1, 2 * i - 1), 2 * i)
+    for _ in range(N):
+        print(randint(1, N), randint(1, N))