regular backup

src/bin/a.cc

@@ -540,29 +540,24 @@ void prep() {
 
 // __attribute__((target("popcnt")))
 void solve() {
-    int n;
+    read(int, n);
-    unordered_set<string> oc;
+    unordered_map<string, int> mp;
-    while (1) {
+    for (int i = 0; i < n; ++i) {
-        read(string, s);
+        read(int, m);
-        if (s[0] >= '0' and s[0] <= '9') {
+        while (m--) {
-            n = stoi(s);
+            read(string, s);
-            break;
+            mp[s] += 1;
         }
-        string nw;
+    }
-        for (auto&& c : s) {
+    vector<string> res;
-            if (c >= 'A' and c <= 'Z') {
+    for (auto&& [x, c] : mp) {
-                nw += c - 'A' + 'a';
+        if (c == n) {
-            } else if (c >= 'a' and c <= 'z') {
+            res.emplace_back(x);
-                nw += c;
-            }
         }
-        oc.emplace(nw);
     }
-    while (n--) {
+    sort(res.begin(), res.end());
-        read(string, s);
+    cout << res.size() << '\n';
-        oc.erase(s);
+    putvec_eol(res);
-    }
-    cout << oc.size() << '\n';
 }
 
 int main() {

src/bin/b.cc

@@ -535,14 +535,29 @@ void dump() {}
 
 void dump_ignore() {}
 
-void prep() {
+constexpr int N = 5e5 + 10;
-}
+ll fact[N], factrev[N + 1], s[N + 1];
+
+void prep() {}
+
 
 // __attribute__((target("popcnt")))
 void solve() {
-    read(string, s);
+    read(int, n, k);
-    string s1(s.rbegin(), s.rend());
+    auto work = [&n] (int k) -> ll {
-    cout << s1 << '\n';
+        vector<ll> dp(n + 1);
+        dp[0] = 1;
+        for (int i = 1; i <= n; ++i) {
+            dp[i] = dp[i - 3];
+            if (i - k - 1 >= 0) {
+                dp[i] -= dp[i - k - 1];
+            }
+            dp[i] = mod(dp[i], PRIME);
+            dp[i] = (dp[i] + dp[i - 1]) % PRIME;
+        }
+        return mod(dp[n] - dp[n - 1], PRIME);
+    };
+    ll cnt = mod(work(k) - work(k - 1), PRIME);
 }
 
 int main() {

src/bin/c.cc

@@ -538,50 +538,186 @@ void dump_ignore() {}
 void prep() {
 }
 
-// __attribute__((target("popcnt")))
+struct SA {
-void solve() {
+    int n;
-    read(int, n);
+    std::vector<int> sa, rk, lc;
-    vector<int> a(n + 1);
+    vector<vector<int>> st;
-    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());
+    SA(std::string s) {
-    sort(ps.begin() + 1, ps.end());
+        n = s.size();
-    for (int i = 1; i <= n; ++i) {
+        sa.resize(n);
-        if (ps[i] > a[i]) {
+        lc.resize(n - 1);
-            cout << "NO\n";
+        rk.resize(n);
-            return;
+        std::iota(sa.begin(), sa.end(), 0);
+        sort_by_key(sa.begin(), sa.end(), expr(s[i], int i));
+        rk[sa[0]] = 0;
+        for (int i = 1; i < n; i++) {
+            rk[sa[i]] = rk[sa[i - 1]] + (s[sa[i]] != s[sa[i - 1]]);
+        }
+        int k = 1;
+        std::vector<int> tmp, cnt(n);
+        tmp.reserve(n);
+        while (rk[sa[n - 1]] < n - 1) {
+            tmp.clear();
+            for (int i = 0; i < k; i++) {
+                tmp.push_back(n - k + i);
+            }
+            for (auto i : sa) {
+                if (i >= k) {
+                    tmp.push_back(i - k);
+                }
+            }
+            std::fill(cnt.begin(), cnt.end(), 0);
+            for (int i = 0; i < n; i++) {
+                cnt[rk[i]]++;
+            }
+            for (int i = 1; i < n; i++) {
+                cnt[i] += cnt[i - 1];
+            }
+            for (int i = n - 1; i >= 0; i--) {
+                sa[--cnt[rk[tmp[i]]]] = tmp[i];
+            }
+            std::swap(rk, tmp);
+            rk[sa[0]] = 0;
+            for (int i = 1; i < n; i++) {
+                rk[sa[i]] = rk[sa[i - 1]] + (tmp[sa[i - 1]] < tmp[sa[i]] || sa[i - 1] + k == n || tmp[sa[i - 1] + k] < tmp[sa[i] + k]);
+            }
+            k *= 2;
+        }
+        for (int i = 0, j = 0; i < n; i++) {
+            if (rk[i] == 0) {
+                j = 0;
+            } else {
+                for (j -= j > 0; i + j < n && sa[rk[i] - 1] + j < n && s[i + j] == s[sa[rk[i] - 1] + j]; ) {
+                    j++;
+                }
+                lc[rk[i] - 1] = j;
+            }
+        }
+        int m = lc.size();
+        int lgm = lg2(m);
+        st = vector(lgm + 1, vector<int>(m));
+        st[0] = lc;
+        for (int j = 0; j < lgm; j++) {
+            for (int i = 0; i + (2 << j) <= m; i++) {
+                st[j + 1][i] = std::min(st[j][i], st[j][i + (1 << j)]);
+            }
         }
     }
-    cout << "YES\n";
+
+    int rmq(int l, int r) {
+        int k = lg2(r - l);
+        return std::min(st[k][l], st[k][r - (1 << k)]);
+    }
+
+    __attribute__((target("lzcnt")))
+    int lcp(int i, int j) {
+        if (i == j || i == n || j == n) {
+            return std::min(n - i, n - j);
+        }
+        int a = rk[i];
+        int b = rk[j];
+        if (a > b) {
+            std::swap(a, b);
+        }
+        return std::min({n - i, n - j, rmq(a, b)});
+    }
+
+    int lcs(int i, int j) {
+        if (i == j || i == 0 || j == 0) {
+            return std::min(i, j);
+        }
+        int a = rk[n + n - i];
+        int b = rk[n + n - j];
+        if (a > b) {
+            std::swap(a, b);
+        }
+        return std::min({i, j, rmq(a, b)});
+    }
+};
+
+
+// __attribute__((target("popcnt")))
+void solve() {
+    read(int, m);
+    read(string, s, t);
+    string p = s + '#' + t;
+    SA sa(p);
+    // vector<int> pp = calc_next(p);
+    vector<ll> pw(5 * m + 1);
+    pw[0] = 1;
+    for (int i = 1; i <= 5 * m; ++i) {
+        pw[i] = pw[i - 1] * 26 % PRIME;
+    }
+    vector<int> ps = calc_z(s);
+    // debug(ps);
+    vector<int> pt = calc_z(t);
+    int n1 = s.size(), n2 = t.size();
+    for (int i = 1; i <= m; ++i) {
+        int tot = 2 * i + n1 + n2;
+        if (tot % 2) {
+            cout << "0 ";
+            continue;
+        }
+        int half = tot / 2;
+        int left = i;
+        int right = i + n1 - 1;
+        int l2 = i + n1 + i;
+
+        int f = 1;
+
+        if (n2 > half) {
+            if (pt[half] >= n2 - half) {
+                ;;
+            } else {
+                f = 0;
+            }
+        }
+        // deb(n1, half, ps[half]);
+        if (n1 > half) {
+            if (ps[half] >= n1 - half) {
+                ;;
+            } else {
+                f = 0;
+            }
+        }
+
+        int cs = max(l2 - half, left), ce = min(half - 1, right);
+        if (cs <= ce) {
+            int ds = cs + half - l2, de = ce + half - l2;
+            cs -= left, ce -= left;
+            if (sa.lcp(cs, n1 + 1 + ds) >= ce - cs + 1) {
+                ;;
+            } else {
+                f = 0;
+            }
+        }
+
+        if (f == 0) {
+            cout << 0 << ' ';
+            continue;
+        }
+
+        vector<pii> segs;
+        segs.emplace_back(max(0, l2 - half), half - 1);
+        segs.emplace_back(left, min(half - 1, right));
+        if (right >= half) {
+            segs.emplace_back(0, right - half);
+        }
+        sort(segs.begin(), segs.end());
+        int rr = -1;
+        int rem = 0;
+        for (auto&& [l, r] : segs) {
+            if (l > rr) {
+                rem += l - rr - 1;
+            }
+            chmax(rr, r);
+        }
+        rem += half - 1 - rr;
+
+        // debug(rem);
+        cout << pw[rem] << ' ';
+    }
 }
 
 int main() {

src/bin/cf-771d.cc

@@ -0,0 +1,612 @@
+// #pragma GCC target("popcnt,lzcnt,abm,bmi,bmi2")
+#pragma GCC optimize("Ofast,unroll-loops")
+/************* This code requires C++17. ***************/
+
+#include<bits/stdc++.h>
+using namespace std;
+
+/* macro helpers */
+#define __NARGS(...) std::tuple_size<decltype(std::make_tuple(__VA_ARGS__))>::value
+#define __DECOMPOSE_S(a, x) auto x = a;
+#define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
+constexpr void __() {}
+#define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
+#define __as_typeof(container) remove_reference<decltype(container)>::type
+template <typename T> struct argument_type;
+template <typename T, typename U> struct argument_type<T(U)> { using type = U; };
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = __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);
+    for (int i = 0; i < n; ++i) {
+        read(char, c);
+        if (c == 'V') {
+            a[i] = 0;
+        } else if (c == 'K') {
+            a[i] = 1;
+        } else {
+            a[i] = 2;
+        }
+    }
+    // vector dp(n, vector(n, vector(n + 1, vector<int>(n + 1, INF))));
+    vector<vector<vector<vector<vector<int>>>>> dp(n);
+    for (int i = 0; i < n; ++i) {
+        dp[i].resize(n - i);
+        for (int j = 0; j < n - i; ++j) {
+            dp[i][j].resize(j + 2);
+            for (int k = 0; k <= j + 1; ++k) {
+                dp[i][j][k].resize(j + 1 - k + 1);
+                for (int l = 0; l <= j + 1 - k; ++l) {
+                    dp[i][j][k][l].resize((j + 1) / 2 + 1);
+                }
+            }
+        }
+    }
+    for (int i = n - 1; ~i; --i) {
+        for (int j = i; j < n; ++j) {
+            if (i == j) {
+                dp[i][0][a[i] == 0][a[i] == 1][0] = 0;
+            } else {
+                if (a[i] == 0) {
+                    for (int k = 0; k <= j - i; ++k) {
+                        for (int l = 0; l <= j - i - k; ++l) {
+                            for (int m = 0; m <= (j - i) / 2; ++m) {
+                                
+                            }
+                        }
+                    }
+                }
+            }
+        }
+    }
+}
+
+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-798d.cc

@@ -0,0 +1,594 @@
+// #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);
+    readvec(int, a, n);
+    readvec(int, b, n);
+    ll sa = accumulate(a.begin(), a.end(), ll(0));
+    ll sb = accumulate(b.begin(), b.end(), ll(0));
+    auto check = [&] (const vector<int>& idx) {
+        ll ca = 0, cb = 0;
+        for (auto&& i : idx) {
+            ca += a[i];
+            cb += b[i];
+        }
+        return 2 * ca > sa and 2 * cb > sb;
+    };
+    vector<int> perm(n);
+    iota(perm.begin(), perm.end(), 0);
+    while (1) {
+        shuffle(perm.begin(), perm.end(), rd);
+        vector<int> curr(perm.begin(), perm.begin() + n / 2 + 1);
+        if (check(curr)) {
+            cout << n / 2 + 1 << '\n';
+            transform(curr.begin(), curr.end(), curr.begin(), expr(x + 1, auto&& x));
+            putvec(curr);
+            break;
+        }
+    }
+}
+
+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 b33b8ed3f4e64e62b354eb3acf0f92ec93a41f75
+Subproject commit dfd419ac9502baf7d916ecbca18ae11521be72d9

src/bin/h.cc

@@ -538,40 +538,71 @@ 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 long double 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 long double slope(const point& a, const point& b) {
+        if (a.x == b.x) return INFLL;
+        return (long double)(a.y - b.y) / (a.x - b.x);
+    }
+
+    // distance from point `a` to line `l--r`
+    static inline long double dist(const point& a, const point& l, const point& r) {
+        return area(a, l, r) * 2 / (l - r).norm();
+    }
+
+    static inline long double area(const point& a, const point& b, const point& c) {
+        return abs((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;
+    }
+};
+
 // __attribute__((target("popcnt")))
 void solve() {
-    read(int, n, q);
+    using point = point<long double>;
-    readvec(int, a, n);
+    read(int, xa, ya);
-    read(int, _);
+    deb(xa, ya);
-    int prod = 1;
+    point a(xa, ya);
-    for (auto&& x : a) prod *= x;
+    read(point, b);
-    if (prod == 0) {
+    read(point, c, d);
-        cout << "Yes\n";
+    point e((a.x + b.x) / ld(2), (a.y + b.y) / ld(2));
-    } else {
+    cout << fixed << setprecision(50);
-        constexpr int D = 5;
+    debug(c - e);
-        vector dp(n, vector(2 * D + 1, vector<int>(2)));
+    debug(d - e);
-        dp[0][D + a[0]][a[0] == 1] = 1;
+    debug((long double)point::area(e, c, d));
-        for (int i = 1; i < n; ++i) {
+    cout << (long double)point::dist(e, c, d) << '\n';
-            for (int j = -D; j <= D; ++j) {
-                for (int k = 0; k < 2; ++k) {
-                    // continue
-                    int nw = (k ? 1 : -1) * a[i];
-                    if (j - (k ? 1 : -1) + nw >= -D and j - (k ? 1 : -1) + nw <= D) {
-                        dp[i][j - (k ? 1 : -1) + nw + D][nw == 1] or_eq dp[i - 1][j + D][k];
-                    }
-                    // new segment
-                    if (j + a[i] >= -D and j + a[i] <= D) {
-                        dp[i][j + a[i] + D][a[i] == 1] or_eq dp[i - 1][j + D][k];
-                    }
-                }
-            }
-        }
-        if (dp[n - 1][D][0] or dp[n - 1][D][1]) {
-            cout << "Yes\n";
-        } else {
-            cout << "No\n";
-        }
-    }
 }
 
 int main() {

src/bin/i.cc

@@ -538,238 +538,38 @@ void dump_ignore() {}
 void prep() {
 }
 
-namespace tarjan {
-    struct mutex_cond {
-        int v1; bool cond1;
-        int v2; bool cond2;
-        mutex_cond(int v1, bool cond1, int v2, bool cond2) : v1(v1), cond1(cond1), v2(v2), cond2(cond2) {}
-    };
-    struct inclusive_cond {
-        int v1; bool cond1;
-        int v2; bool cond2;
-        inclusive_cond(int v1, bool cond1, int v2, bool cond2) : v1(v1), cond1(cond1), v2(v2), cond2(cond2) {}
-    };
-    // Returns the mapping between vertices and their affiliated sccs.
-    vector<int> scc(const vector<vector<int>>& ch) {
-        int n = ch.size();
-        int cnt = 0, scn = 0;
-        vector<int> dfn(n), low(n), vis(n), st;
-        vector<int> br(n);
-        auto tarjan = [&] (auto tarjan, int v) -> void {
-            dfn[v]=low[v]=++cnt;
-            st.push_back(v);
-            vis[v]=1;
-            for(const auto&u:ch[v])
-                if(!dfn[u]) tarjan(tarjan, u),low[v]=min(low[v],low[u]);
-                else if(vis[u])low[v]=min(low[v],dfn[u]);
-            if(dfn[v]==low[v]){
-                ++scn;
-                int u;
-                do u=st.back(), st.pop_back(),vis[u]=0,br[u]=scn; while(u!=v);
-            }
-        };
-        for (int i = 0; i < n; ++i) {
-            if (!dfn[i]) {
-                tarjan(tarjan, i);
-            }
-        }
-        return br;
-    }
-    // This method can eliminate redundant edges or self-loops
-    vector<vector<int>> build_scc(const vector<vector<int>>& ch) {
-        int n = ch.size();
-        auto br = scc(ch);
-        int cnt = *max_element(br.begin(), br.end());
-        vector<unordered_set<int, safe_hash>> rb(cnt + 1);
-        for (int i = 0; i < n; ++i) {
-            for (auto&& u : ch[i]) {
-                if (br[i] != br[u]) rb[br[i]].emplace(br[u]);
-            }
-        }
-        vector<vector<int>> res(cnt + 1);
-        for (int i = 1; i <= cnt; ++i) {
-            res[i] = vector<int>(rb[i].begin(), rb[i].end());
-        }
-        return res;
-    }
-
-    // This method can eliminate redundant edges or self-loops
-    // return form: (scc size, children of scc)
-    vector<pair<size_t, vector<int>>> build_scc_with_size(const vector<vector<int>>& ch) {
-        int n = ch.size();
-        auto br = scc(ch);
-        int cnt = *max_element(br.begin(), br.end());
-        vector<unordered_set<int, safe_hash>> rb(cnt + 1);
-        for (int i = 0; i < n; ++i) {
-            for (auto&& u : ch[i]) {
-                if (br[i] != br[u]) rb[br[i]].emplace(br[u]);
-            }
-        }
-        vector<pair<size_t, vector<int>>> res(cnt + 1);
-        for (int i = 1; i <= cnt; ++i) {
-            res[i].second = vector<int>(rb[i].begin(), rb[i].end());
-        }
-        for (int i = 1; i <= n; ++i) {
-            res[br[i]].first += 1;
-        }
-        return res;
-    }
-    // indices start from 1, result has `n` items
-    optional<vector<bool>> solve_twosat(int n, const vector<mutex_cond>& conditions) {
-        vector<vector<int>> ch(2 * n + 1);
-        for (auto&& [v1, cond1, v2, cond2] : conditions) {
-            ch[(1 ^ cond1) * n + v1].emplace_back(cond2 * n + v2);
-            ch[(1 ^ cond2) * n + v2].emplace_back(cond1 * n + v1);
-        }
-        auto sccno = scc(ch);
-        for (int i = 1; i <= n; ++i) {
-            if (sccno[i] == sccno[i + n]) {
-                return nullopt;
-            }
-        }
-        vector<bool> res;
-        for (int i = 1; i <= n; ++i) {
-            if (sccno[i] < sccno[i + n]) {
-                res.emplace_back(false);
-            } else {
-                res.emplace_back(true);
-            }
-        }
-        return res;
-    };
-    // indices start from 1, result has `n` items
-    optional<vector<bool>> solve_twosat(int n, const vector<inclusive_cond>& conditions) {
-        vector<mutex_cond> trans_conds;
-        for (auto&& [v1, cond1, v2, cond2] : conditions) {
-            trans_conds.emplace_back(v1, cond1, v2, not cond2);
-        }
-        return solve_twosat(n, trans_conds);
-    }
-
-    // Returns if each vertex is a cut vertex
-    // All indices start from 1
-    vector<int> cut_v(const vector<vector<int>>& ch) {
-        int n = ch.size() - 1;
-        vector<bool> vis(n + 1);
-        vector<int> low(n + 1), dfn(n + 1), flag(n + 1);
-        int cnt = 0;
-        auto dfs = [&] (auto dfs, int v, int pa) -> void {
-            vis[v] = 1;
-            low[v] = dfn[v] = ++cnt;
-            int child = 0;
-            for (auto&& u : ch[v]) {
-                if (not vis[u]) {
-                    ++child;
-                    dfs(dfs, u, v);
-                    low[v] = min(low[v], low[u]);
-                    if (pa != v and low[u] >= dfn[v] and not flag[v]) {
-                        flag[v] = 1;
-                    }
-                } else if (u != pa) {
-                    low[v] = min(low[v], dfn[u]);
-                }
-            }
-            if (pa == v and child >= 2 and not flag[v]) {
-                flag[v] = 1;
-            }
-        };
-        for (int i = 1; i <= n; ++i) {
-            if (not dfn[i]) {
-                dfs(dfs, i, 0);
-            }
-        }
-        return flag;
-    }
-}
-
-class quick_union {
-private:
-    vector<size_t> c, sz;
-public:
-    quick_union(size_t n) : c(n), sz(n) {
-        iota(c.begin(), c.end(), 0);
-        sz.assign(n, 1);
-    }
-    
-    size_t query(size_t i) {
-        if (c[i] != i) c[i] = query(c[i]);
-        return c[i];
-    }
-    
-    void merge(size_t i, size_t j) {
-        if (connected(i, j)) return;
-        sz[query(j)] += sz[query(i)];
-        c[query(i)] = query(j);
-    }
-
-    bool connected(size_t i, size_t j) {
-        return query(i) == query(j);
-    }
-
-    size_t query_size(size_t i) {
-        return sz[query(i)];
-    }
-};
-
-
 // __attribute__((target("popcnt")))
 void solve() {
     read(int, n);
-    adj(ch, n);
+    vector<int> res;
-    quick_union qu(n + 1);
+    int start = 1;
-    vector<pii> raw;
+    if (n % 8 == 1) {
-    for (int i = 1; i <= n; ++i) {
+        res.emplace_back(1);
-        read(int, j);
+        start = 2;
-        Edge(ch, i, j);
+    } else if (n % 8 == 2) {
-        raw.emplace_back(i, j);
+        res = { 1, 2 };
-        qu.merge(i, j);
+        start = 3;
+    } else if (n % 8 == 3) {
+        res = { 1, 2, 3 };
+        start = 4;
+    } else if (n % 8 == 4) {
+        res = { 1, 2, 3, 4 };
+        start = 5;
+    } else if (n % 8 == 5) {
+        res = { 5, 2, 1, 4, 3 };
+        start = 6;
+    } else if (n % 8 == 6) {
+        res = { 5, 2, 3, 6, 1, 4 };
+        start = 7;
+    } else if (n % 8 == 7) {
+        res = { 1, 4, 7, 2, 3, 6, 5 };
+        start = 8;
     }
-
+    for (int i = start; i + 7 <= n; i += 8) {
-    vector<vector<int>> cc(n + 1);;
+        vector<int> curr = { i + 2, i + 7, i + 4, i + 1, i + 6, i + 3, i, i + 5 };
-    for (int i = 1; i <= n; ++i) {
+        res.insert(res.end(), curr.begin(), curr.end());
-        cc[qu.query(i)].emplace_back(i);
     }
-
+    putvec(res);
-    vector<vector<vector<int>>> nch(n + 1);
-    for (int i = 1; i <= n; ++i) {
-        nch[i].assign(cc[i].size() + 1, {});
-    }
-    int cnt = 0;
-
-    vector<int> mapped(n + 1);
-
-    for (auto&& ns : cc) {
-        if (ns.empty()) continue;
-        cnt += 1;
-        int m = ns.size();
-        for (int i = 0; i < m; ++i) {
-            mapped[ns[i]] = i + 1;
-        }
-    }
-
-    for (auto&& [u, v] : raw) {
-        Edge(nch[qu.query(u)], mapped[u], mapped[v]);
-    }
-
-    int res = cnt - 1;
-
-    for (int i = 1; i <= n; ++i) {
-        if (cc[i].empty()) continue;
-        auto scc = tarjan::scc(nch[i]);
-        int f = 1;
-        for (int j = 2; j <= cc[i].size(); ++j) {
-            if (scc[j] != scc[1]) {
-                f = 0;
-                break;
-            }
-        }
-        if (f == 0) {
-            res += 1;
-        }
-    }
-    
-    cout << res << '\n';
-    
 }
 
 int main() {

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
 
@@ -538,18 +538,129 @@ void dump_ignore() {}
 void prep() {
 }
 
+template<typename T>
+struct BIT {
+    int n;
+    vector<T> c;
+    BIT(size_t n) : n(n), c(n + 1) {}
+    void add(size_t i, const T& k) {
+        while (i <= n) {
+            c[i] += k;
+            i += lowbit(i);
+        }
+    }
+    T getsum(size_t i) {
+        T res = {};
+        while (i) {
+            res += c[i];
+            i -= lowbit(i);
+        }
+        return res;
+    }
+};
+
+
 // __attribute__((target("popcnt")))
 void solve() {
-    read(int, n);
+    read(int, n, m);
-    read(int, a, b, c);
+    vector a(n, vector<int>(m));
-    int res = 0;
+    for (int i = 0; i < n; ++i) {
-    readvec(pii, s, n);
+        for (int j = 0; j < m; ++j) {
-    read(int, d);
+            cin >> a[i][j];
-    for (auto&& [x, y] : s) {
+        }
-        if (x + y >= c) continue;
+    }
-        if (min(a, x + d) + y >= c) res += 1;
+    vector<pii> rows;
+    vector<vector<int>> cols;
+    bool row_sorted = 1, col_sorted = 1;
+    int prev_mx = -1;
+    for (int i = 0; i < n; ++i) {
+        int mn = INF, mx = -INF;
+        for (int j = 0; j < m; ++j) {
+            chmin(mn, a[i][j]);
+            chmax(mx, a[i][j]);
+        }
+        vector<int> idx(m);
+        iota(idx.begin(), idx.end(), 0);
+        sort_by_key(idx.begin(), idx.end(), expr(a[i][j], auto&& j));
+        cols.emplace_back(idx);
+        rows.emplace_back(mn, mx);
+        if (mn > prev_mx) {
+            ;;
+        } else {
+            row_sorted = 0;
+        }
+        prev_mx = mx;
+    }
+    for (int i = 1; i < m; ++i) {
+        if (cols[0][i] > cols[0][i - 1]) {
+            ;;
+        } else {
+            col_sorted = 0;
+        }
+    }
+    vector<int> rowidx(n), colidx(m);
+    iota(rowidx.begin(), rowidx.end(), 0);
+    iota(colidx.begin(), colidx.end(), 0);
+    sort_by_key(rowidx.begin(), rowidx.end(), expr(rows[i], auto i));
+    sort_by_key(colidx.begin(), colidx.end(), expr(a[0][i], auto i));
+    for (int i = 1; i < n; ++i) {
+        if (rows[rowidx[i]].first < rows[rowidx[i - 1]].second or cols[rowidx[i]] != cols[rowidx[i - 1]]) {
+            cout << "NSFW\n";
+            return;
+        }
+    }
+    int rowinv = 0, colinv = 0;
+    {
+        vector<bool> vis(n);
+        int cnt;
+        {
+            auto dfs = [&] (auto dfs, int v) -> void {
+                vis[v] = 1;
+                cnt += 1;
+                if (not vis[rowidx[v]]) dfs(dfs, rowidx[v]);
+            };
+            for (int i = 0; i < n; ++i) {
+                if (not vis[i]) {
+                    cnt = 0;
+                    dfs(dfs, i);
+                    rowinv += cnt - 1;
+                }
+            }
+        }
+    }
+    {
+        vector<bool> vis(m);
+        int cnt;
+        {
+            auto dfs = [&] (auto dfs, int v) -> void {
+                vis[v] = 1;
+                cnt += 1;
+                if (not vis[colidx[v]]) dfs(dfs, colidx[v]);
+            };
+            for (int i = 0; i < m; ++i) {
+                if (not vis[i]) {
+                    cnt = 0;
+                    dfs(dfs, i);
+                    colinv += cnt - 1;
+                }
+            }
+        }
+    }
+    if (rowinv % 2 == colinv % 2) {
+        // fox loses
+        if (n == 2) {
+            cout << "CAT\n";
+        } else {
+            cout << "NSFW\n";
+        }
+    } else {
+        // cat loses
+        if (m == 2 or rowinv == 1 and colinv == 0) {
+            cout << "FOX\n";
+        } else {
+            cout << "NSFW\n";
+        }
     }
-    cout << res<< '\n';
 }
 
 int main() {

src/bin/nc-57362a.cc

@@ -0,0 +1,587 @@
+// #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);
+    unordered_map<string, int> mp;
+    for (int i = 0; i < n; ++i) {
+        read(int, m);
+        while (m--) {
+            read(string, s);
+            mp[s] += 1;
+        }
+    }
+    vector<string> res;
+    for (auto&& [x, c] : mp) {
+        if (c == n) {
+            res.emplace_back(x);
+        }
+    }
+    sort(res.begin(), res.end());
+    cout << res.size() << '\n';
+    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/nc-57362i.cc

@@ -0,0 +1,747 @@
+// #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() {
+}
+
+struct SA {
+    int n;
+    std::vector<int> sa, rk, lc;
+    vector<vector<int>> st;
+
+    SA(std::string s) {
+        n = s.size();
+        sa.resize(n);
+        lc.resize(n - 1);
+        rk.resize(n);
+        std::iota(sa.begin(), sa.end(), 0);
+        sort_by_key(sa.begin(), sa.end(), expr(s[i], int i));
+        rk[sa[0]] = 0;
+        for (int i = 1; i < n; i++) {
+            rk[sa[i]] = rk[sa[i - 1]] + (s[sa[i]] != s[sa[i - 1]]);
+        }
+        int k = 1;
+        std::vector<int> tmp, cnt(n);
+        tmp.reserve(n);
+        while (rk[sa[n - 1]] < n - 1) {
+            tmp.clear();
+            for (int i = 0; i < k; i++) {
+                tmp.push_back(n - k + i);
+            }
+            for (auto i : sa) {
+                if (i >= k) {
+                    tmp.push_back(i - k);
+                }
+            }
+            std::fill(cnt.begin(), cnt.end(), 0);
+            for (int i = 0; i < n; i++) {
+                cnt[rk[i]]++;
+            }
+            for (int i = 1; i < n; i++) {
+                cnt[i] += cnt[i - 1];
+            }
+            for (int i = n - 1; i >= 0; i--) {
+                sa[--cnt[rk[tmp[i]]]] = tmp[i];
+            }
+            std::swap(rk, tmp);
+            rk[sa[0]] = 0;
+            for (int i = 1; i < n; i++) {
+                rk[sa[i]] = rk[sa[i - 1]] + (tmp[sa[i - 1]] < tmp[sa[i]] || sa[i - 1] + k == n || tmp[sa[i - 1] + k] < tmp[sa[i] + k]);
+            }
+            k *= 2;
+        }
+        for (int i = 0, j = 0; i < n; i++) {
+            if (rk[i] == 0) {
+                j = 0;
+            } else {
+                for (j -= j > 0; i + j < n && sa[rk[i] - 1] + j < n && s[i + j] == s[sa[rk[i] - 1] + j]; ) {
+                    j++;
+                }
+                lc[rk[i] - 1] = j;
+            }
+        }
+        int m = lc.size();
+        int lgm = lg2(m);
+        st = vector(lgm + 1, vector<int>(m));
+        st[0] = lc;
+        for (int j = 0; j < lgm; j++) {
+            for (int i = 0; i + (2 << j) <= m; i++) {
+                st[j + 1][i] = std::min(st[j][i], st[j][i + (1 << j)]);
+            }
+        }
+    }
+
+    int rmq(int l, int r) {
+        int k = lg2(r - l);
+        return std::min(st[k][l], st[k][r - (1 << k)]);
+    }
+
+    __attribute__((target("lzcnt")))
+    int lcp(int i, int j) {
+        if (i == j || i == n || j == n) {
+            return std::min(n - i, n - j);
+        }
+        int a = rk[i];
+        int b = rk[j];
+        if (a > b) {
+            std::swap(a, b);
+        }
+        return std::min({n - i, n - j, rmq(a, b)});
+    }
+
+    int lcs(int i, int j) {
+        if (i == j || i == 0 || j == 0) {
+            return std::min(i, j);
+        }
+        int a = rk[n + n - i];
+        int b = rk[n + n - j];
+        if (a > b) {
+            std::swap(a, b);
+        }
+        return std::min({i, j, rmq(a, b)});
+    }
+};
+
+
+// __attribute__((target("popcnt")))
+void solve() {
+    read(int, m);
+    read(string, s, t);
+    string p = s + '#' + t;
+    SA sa(p);
+    // vector<int> pp = calc_next(p);
+    vector<ll> pw(5 * m + 1);
+    pw[0] = 1;
+    for (int i = 1; i <= 5 * m; ++i) {
+        pw[i] = pw[i - 1] * 26 % PRIME;
+    }
+    vector<int> ps = calc_z(s);
+    // debug(ps);
+    vector<int> pt = calc_z(t);
+    int n1 = s.size(), n2 = t.size();
+    for (int i = 1; i <= m; ++i) {
+        int tot = 2 * i + n1 + n2;
+        if (tot % 2) {
+            cout << "0 ";
+            continue;
+        }
+        int half = tot / 2;
+        int left = i;
+        int right = i + n1 - 1;
+        int l2 = i + n1 + i;
+
+        int f = 1;
+
+        if (n2 > half) {
+            if (pt[half] >= n2 - half) {
+                ;;
+            } else {
+                f = 0;
+            }
+        }
+        // deb(n1, half, ps[half]);
+        if (n1 > half) {
+            if (ps[half] >= n1 - half) {
+                ;;
+            } else {
+                f = 0;
+            }
+        }
+
+        int cs = max(l2 - half, left), ce = min(half - 1, right);
+        if (cs <= ce) {
+            int ds = cs + half - l2, de = ce + half - l2;
+            cs -= left, ce -= left;
+            if (sa.lcp(cs, n1 + 1 + ds) >= ce - cs + 1) {
+                ;;
+            } else {
+                f = 0;
+            }
+        }
+
+        if (f == 0) {
+            cout << 0 << ' ';
+            continue;
+        }
+
+        vector<pii> segs;
+        segs.emplace_back(max(0, l2 - half), half - 1);
+        segs.emplace_back(left, min(half - 1, right));
+        if (right >= half) {
+            segs.emplace_back(0, right - half);
+        }
+        sort(segs.begin(), segs.end());
+        int rr = -1;
+        int rem = 0;
+        for (auto&& [l, r] : segs) {
+            if (l > rr) {
+                rem += l - rr - 1;
+            }
+            chmax(rr, r);
+        }
+        rem += half - 1 - rr;
+
+        // debug(rem);
+        cout << pw[rem] << ' ';
+    }
+}
+
+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/nc-57362j.cc

@@ -0,0 +1,599 @@
+// #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> res;
+    int start = 1;
+    if (n % 8 == 1) {
+        res.emplace_back(1);
+        start = 2;
+    } else if (n % 8 == 2) {
+        res = { 1, 2 };
+        start = 3;
+    } else if (n % 8 == 3) {
+        res = { 1, 2, 3 };
+        start = 4;
+    } else if (n % 8 == 4) {
+        res = { 1, 2, 3, 4 };
+        start = 5;
+    } else if (n % 8 == 5) {
+        res = { 5, 2, 1, 4, 3 };
+        start = 6;
+    } else if (n % 8 == 6) {
+        res = { 5, 2, 3, 6, 1, 4 };
+        start = 7;
+    } else if (n % 8 == 7) {
+        res = { 1, 4, 7, 2, 3, 6, 5 };
+        start = 8;
+    }
+    for (int i = start; i + 7 <= n; i += 8) {
+        vector<int> curr = { i + 2, i + 7, i + 4, i + 1, i + 6, i + 3, i, i + 5 };
+        res.insert(res.end(), curr.begin(), curr.end());
+    }
+    putvec(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/nc-57362k.cc

@@ -0,0 +1,690 @@
+// #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 BIT {
+    int n;
+    vector<T> c;
+    BIT(size_t n) : n(n), c(n + 1) {}
+    void add(size_t i, const T& k) {
+        while (i <= n) {
+            c[i] += k;
+            i += lowbit(i);
+        }
+    }
+    T getsum(size_t i) {
+        T res = {};
+        while (i) {
+            res += c[i];
+            i -= lowbit(i);
+        }
+        return res;
+    }
+};
+
+
+// __attribute__((target("popcnt")))
+void solve() {
+    read(int, n, m);
+    vector a(n, vector<int>(m));
+    for (int i = 0; i < n; ++i) {
+        for (int j = 0; j < m; ++j) {
+            cin >> a[i][j];
+        }
+    }
+    vector<pii> rows;
+    vector<vector<int>> cols;
+    bool row_sorted = 1, col_sorted = 1;
+    int prev_mx = -1;
+    for (int i = 0; i < n; ++i) {
+        int mn = INF, mx = -INF;
+        for (int j = 0; j < m; ++j) {
+            chmin(mn, a[i][j]);
+            chmax(mx, a[i][j]);
+        }
+        vector<int> idx(m);
+        iota(idx.begin(), idx.end(), 0);
+        sort_by_key(idx.begin(), idx.end(), expr(a[i][j], auto&& j));
+        cols.emplace_back(idx);
+        rows.emplace_back(mn, mx);
+        if (mn > prev_mx) {
+            ;;
+        } else {
+            row_sorted = 0;
+        }
+        prev_mx = mx;
+    }
+    for (int i = 1; i < m; ++i) {
+        if (cols[0][i] > cols[0][i - 1]) {
+            ;;
+        } else {
+            col_sorted = 0;
+        }
+    }
+    vector<int> rowidx(n), colidx(m);
+    iota(rowidx.begin(), rowidx.end(), 0);
+    iota(colidx.begin(), colidx.end(), 0);
+    sort_by_key(rowidx.begin(), rowidx.end(), expr(rows[i], auto i));
+    sort_by_key(colidx.begin(), colidx.end(), expr(a[0][i], auto i));
+    for (int i = 1; i < n; ++i) {
+        if (rows[rowidx[i]].first < rows[rowidx[i - 1]].second or cols[rowidx[i]] != cols[rowidx[i - 1]]) {
+            cout << "NSFW\n";
+            return;
+        }
+    }
+    int rowinv = 0, colinv = 0;
+    {
+        vector<bool> vis(n);
+        int cnt;
+        {
+            auto dfs = [&] (auto dfs, int v) -> void {
+                vis[v] = 1;
+                cnt += 1;
+                if (not vis[rowidx[v]]) dfs(dfs, rowidx[v]);
+            };
+            for (int i = 0; i < n; ++i) {
+                if (not vis[i]) {
+                    cnt = 0;
+                    dfs(dfs, i);
+                    rowinv += cnt - 1;
+                }
+            }
+        }
+    }
+    {
+        vector<bool> vis(m);
+        int cnt;
+        {
+            auto dfs = [&] (auto dfs, int v) -> void {
+                vis[v] = 1;
+                cnt += 1;
+                if (not vis[colidx[v]]) dfs(dfs, colidx[v]);
+            };
+            for (int i = 0; i < m; ++i) {
+                if (not vis[i]) {
+                    cnt = 0;
+                    dfs(dfs, i);
+                    colinv += cnt - 1;
+                }
+            }
+        }
+    }
+    if (rowinv % 2 == colinv % 2) {
+        // fox loses
+        if (n == 2) {
+            cout << "CAT\n";
+        } else {
+            cout << "NSFW\n";
+        }
+    } else {
+        // cat loses
+        if (m == 2 or rowinv == 1 and colinv == 0) {
+            cout << "FOX\n";
+        } else {
+            cout << "NSFW\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/std.in

@@ -1,7 +1,4 @@
-5 2 3
+6
-2 4
+bbabbababbab
-4 3
+bab
-4 2
-4 1
-2 5
 

src/bin/test.py

@@ -8,10 +8,12 @@ import io
 PRIME = 998_244_353
 
 if __name__ == '__main__':
-    N = 100000
+    res = 0
-    M = N // 2
+    N = 75
-    print(N, M, N)
+    for i in range(N):
-    for i in range(1, M + 1):
+        for j in range(i + 1, N):
-        print(randint(1, 2 * i - 1), 2 * i)
+            for k in range(0, j - i + 1 + 1):
-    for _ in range(N):
+                for l in range(0, j - i + 1 - k + 1):
-        print(randint(1, N), randint(1, N))
+                    for m in range(0, (j - i + 1) // 2 + 1):
+                        res += 1
+    print(res)