backup

src/bin/a.cc

@@ -182,6 +182,22 @@ ll inverse(ll a, ll b) {
     return mod(x, b);
 }
 
+vector<tuple<int, int, ll>> decompose(ll x) {
+    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;
+}
+
 /* string algorithms */
 vector<int> calc_next(string t) {  // pi function of t
   int n = (int)t.length();
@@ -241,21 +257,21 @@ void dump() {}
 void prep() {}
 
 void solve() {
-    read(int, n, m, k);
+    read(int, n);
-    readvec(int, a, n);
+    cout << "0 ";
-    readvec(int, b, m);
+    for (int i = 1; i <= (n - 1) / 2; ++i) {
-    sort(a.begin(), a.end());
+        cout << i << ' ' << -i << ' ';
-    sort(b.begin(), b.end());
-    int res = 0;
-    for (int i = 0; i < n; ++i) {
-        int j = -1;
-        while (j + 1 < m && b[j + 1] + a[i] <= k) ++j;
-        res += j + 1;
     }
-    cout << res << endl;
+    if (n % 2 == 0) {
+        cout << (n - 1) / 2 + 1;
+    }
+    cout << '\n';
 }
 
 int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
     untie, cout.tie(NULL);
     prep();
 #ifdef SINGLE_TEST_CASE

src/bin/b.cc

@@ -1,4 +1,3 @@
-#pragma GCC optimize("Ofast")
 /////////////////////////////////////////////////////////
 /**
  * Useful Macros
@@ -182,6 +181,22 @@ ll inverse(ll a, ll b) {
     return mod(x, b);
 }
 
+vector<tuple<int, int, ll>> decompose(ll x) {
+    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;
+}
+
 /* string algorithms */
 vector<int> calc_next(string t) {  // pi function of t
   int n = (int)t.length();
@@ -242,23 +257,43 @@ void prep() {}
 
 void solve() {
     read(int, n);
-    readvec(ll, a, n);
+    readvec(ll, a, n + 1);
-    for (int i = 1; i < n - 1; ++i) {
+    vector<ll> ps(n + 2);
-        if (a[i - 1] < 0) {
+    for (int i = 1; i <= n + 1; ++i) {
-            cout << "NO\n";
+        ps[i] = ps[i - 1] ^ a[i - 1];
-            return;
     }
-        a[i] -= 2 * a[i - 1];
+    vector<int> res(n + 2);
-        a[i + 1] -= a[i - 1];
+    for (int i = 0; i <= n; ++i) {
+        if (ps[i] == a[i]) {
+            res[max(0, n - i - 1)] += 1;
         }
-    if (a[n - 1] != 0 || a[n - 2] != 0) {
+    }
-        cout << "NO\n";
+    int ptr = 0;
+    while (ptr <= n + 1 && res[ptr] == 2) {
+        res[ptr] = 0;
+        res[++ptr] += 1;
+    }
+    int lo = 0;
+    for (int i = n + 1; ~i; --i) {
+        if (res[i] == 0) {
+            if (lo == 0) {
+                ;;
             } else {
-        cout << "YES\n";
+                cout << 0;
             }
+        } else {
+            lo = 1;
+            cout << 1;
+        }
+    }
+    if (lo == 0) cout << 0;
+    cout << '\n';
 }
 
 int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
     untie, cout.tie(NULL);
     prep();
 #ifdef SINGLE_TEST_CASE

src/bin/c.cc

@@ -1,4 +1,3 @@
-#pragma GCC optimize("Ofast")
 /////////////////////////////////////////////////////////
 /**
  * Useful Macros
@@ -182,6 +181,22 @@ ll inverse(ll a, ll b) {
     return mod(x, b);
 }
 
+vector<tuple<int, int, ll>> decompose(ll x) {
+    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;
+}
+
 /* string algorithms */
 vector<int> calc_next(string t) {  // pi function of t
   int n = (int)t.length();
@@ -233,25 +248,66 @@ int period(string s) {  // find the length of shortest recurring period
 }
 /////////////////////////////////////////////////////////
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 512
 
 void dump() {}
 
 void prep() {}
 
-void solve() {
+template<typename T>
-    read(int, n);
+struct BIT {
-    read(string, a);
+    int n;
-    int res = 0;
+    vector<T> c;
-    for (int i = 2; i < n; ++i) {
+    BIT(size_t n) : n(n), c(n + 1) {}
-        if (a[i - 2] == 'p' && a[i - 1] == 'i' && a[i] == 'e') res += 1;
+    void add(size_t i, const T& k) {
-        if (a[i - 2] == 'm' && a[i - 1] == 'a' && a[i] == 'p') res += 1, a[i] = 'x';
+        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;
+    }
+};
+
+void solve() {
+    read(int, n, q);
+    BIT<int> tr(n);
+    for (int i = 1; i <= n; ++i) {
+        read(int, x);
+        tr.add(i, x);
+    }
+    while (q--) {
+        read(int, x, p);
+        tr.add(x, ll(p) - tr.getsum(x) + tr.getsum(x - 1));
+        int l = 0, r = lg2(n);
+        while (l < r) {
+            int mid = l + r >> 1;
+            if (tr.getsum(n - (1 << mid)) == 0) {
+                r = mid;
+            } else {
+                l = mid + 1;
+            }
+        }
+        // debug(l);
+        if (tr.getsum(n - (1 << l)) != 0) {
+            cout << (1 << l + 1) << '\n';
+        } else {
+            cout << (1 << l) << '\n';
+        }
     }
-    cout << res << endl;
 }
 
 int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
     untie, cout.tie(NULL);
     prep();
 #ifdef SINGLE_TEST_CASE

src/bin/cc-128ba.cc

@@ -0,0 +1,292 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (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) decltype(container)::value_type
+
+/* type aliases */
+using ll = int64_t;
+using ull = uint64_t;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+
+/* random */
+
+mt19937 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;
+    }
+};
+
+/* build data structures */
+#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 pa(a) __AS_PROCEDURE(__typeof(a) pa; pa.push_back({}); for (auto&&x : a) pa.push_back(pa.back() + x);)
+#define sa(a) __AS_PROCEDURE(__typeof(a) sa(a.size() + 1); {int n = a.size(); for (int i = n - 1; i >= 0; --i) sa[i] = sa[i + 1] + a[i];};)
+#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 Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+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))
+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(type, ...) __AS_PROCEDURE(type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (int i = 0; i < (n); ++i) cin >> a[i];)
+#define putvec(a) __AS_PROCEDURE(for (auto&& x : a) cout << x << ' '; cout << endl;)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    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;
+}
+
+/* pops */
+#define poptop(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.top(); q.pop();)
+#define popback(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.back(); q.pop_back();)
+#define popfront(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.front();q.pop_front();)
+
+/* math */
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+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) {
+    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;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 512
+
+void dump() {}
+
+void prep() {}
+
+void solve() {
+    read(int, n);
+    readvec(int, a, n);
+    sort(a.begin(), a.end());
+    ll res = 0;
+    for (int i = 0; i < n; ++i) {
+        res += abs(a[i] - i);
+    }
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie, cout.tie(NULL);
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            solve();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cc-128bb.cc

@@ -0,0 +1,314 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (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) decltype(container)::value_type
+
+/* type aliases */
+using ll = int64_t;
+using ull = uint64_t;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+
+/* random */
+
+mt19937 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;
+    }
+};
+
+/* build data structures */
+#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 pa(a) __AS_PROCEDURE(__typeof(a) pa; pa.push_back({}); for (auto&&x : a) pa.push_back(pa.back() + x);)
+#define sa(a) __AS_PROCEDURE(__typeof(a) sa(a.size() + 1); {int n = a.size(); for (int i = n - 1; i >= 0; --i) sa[i] = sa[i + 1] + a[i];};)
+#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 Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+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))
+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(type, ...) __AS_PROCEDURE(type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (int i = 0; i < (n); ++i) cin >> a[i];)
+#define putvec(a) __AS_PROCEDURE(for (auto&& x : a) cout << x << ' '; cout << endl;)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    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;
+}
+
+/* pops */
+#define poptop(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.top(); q.pop();)
+#define popback(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.back(); q.pop_back();)
+#define popfront(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.front();q.pop_front();)
+
+/* math */
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+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) {
+    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;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 512
+
+void dump() {}
+
+void prep() {}
+
+void solve() {
+    read(int, n);
+    readvec(ll, a, n);
+    vector<bool> mark(n);
+    ll x = 0;
+    ll prev = LLONG_MIN;
+    int cb = 0;
+    for (int i = 0; i < n; ++i) {
+        if (a[i] < prev) {
+            if (cb == 1) {
+                cout << "No\n";
+                return;
+            }
+            mark[i] = 1;
+            cb = 1;
+            x = max(x, prev - a[i]);
+        } else {
+            prev = a[i];
+            cb = 0;
+        }
+    }
+    for (int i = 0; i < n; ++i) {
+        if (mark[i]) a[i] += x;
+    }
+    for (int i = 1; i < n; ++i) {
+        if (a[i] < a[i - 1]) {
+            cout << "No\n";
+            return;
+        }
+    }
+    cout << "Yes\n";
+}
+
+int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie, cout.tie(NULL);
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            solve();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cc-128bc.cc

@@ -0,0 +1,298 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (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) decltype(container)::value_type
+
+/* type aliases */
+using ll = int64_t;
+using ull = uint64_t;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+
+/* random */
+
+mt19937 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;
+    }
+};
+
+/* build data structures */
+#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 pa(a) __AS_PROCEDURE(__typeof(a) pa; pa.push_back({}); for (auto&&x : a) pa.push_back(pa.back() + x);)
+#define sa(a) __AS_PROCEDURE(__typeof(a) sa(a.size() + 1); {int n = a.size(); for (int i = n - 1; i >= 0; --i) sa[i] = sa[i + 1] + a[i];};)
+#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 Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+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))
+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(type, ...) __AS_PROCEDURE(type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (int i = 0; i < (n); ++i) cin >> a[i];)
+#define putvec(a) __AS_PROCEDURE(for (auto&& x : a) cout << x << ' '; cout << endl;)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    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;
+}
+
+/* pops */
+#define poptop(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.top(); q.pop();)
+#define popback(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.back(); q.pop_back();)
+#define popfront(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.front();q.pop_front();)
+
+/* math */
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+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) {
+    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;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 512
+
+void dump() {}
+
+void prep() {}
+
+void solve() {
+    read(int, n);
+    read(string, a);
+    int cnt_00 = 0, cnt_11 = 0, cnt_01 = 0, cnt_10 = 0;
+    for (int i = 0; i < n; i += 2) {
+        string curr = a.substr(i, 2);
+        if (curr == "00") cnt_00 += 1;
+        else if (curr == "01") cnt_01 += 1;
+        else if (curr == "10") cnt_10 += 1;
+        else cnt_11 += 1;
+    }
+    int res = 0;
+    res = max(res, cnt_00 * 2 + (!!cnt_01) * 2 + cnt_11 * 2 + min(2, cnt_10));
+    // res = max(res, (!!cnt_10) + cnt_00 * 2 + (!!cnt_01) * 2) + cnt_11 * 2;
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie, cout.tie(NULL);
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            solve();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cc-128bd.cc

@@ -0,0 +1,286 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (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) decltype(container)::value_type
+
+/* type aliases */
+using ll = int64_t;
+using ull = uint64_t;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+
+/* random */
+
+mt19937 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;
+    }
+};
+
+/* build data structures */
+#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 pa(a) __AS_PROCEDURE(__typeof(a) pa; pa.push_back({}); for (auto&&x : a) pa.push_back(pa.back() + x);)
+#define sa(a) __AS_PROCEDURE(__typeof(a) sa(a.size() + 1); {int n = a.size(); for (int i = n - 1; i >= 0; --i) sa[i] = sa[i + 1] + a[i];};)
+#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 Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+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))
+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(type, ...) __AS_PROCEDURE(type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (int i = 0; i < (n); ++i) cin >> a[i];)
+#define putvec(a) __AS_PROCEDURE(for (auto&& x : a) cout << x << ' '; cout << endl;)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    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;
+}
+
+/* pops */
+#define poptop(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.top(); q.pop();)
+#define popback(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.back(); q.pop_back();)
+#define popfront(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.front();q.pop_front();)
+
+/* math */
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+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) {
+    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;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 512
+
+void dump() {}
+
+void prep() {}
+
+void solve() {
+    read(ll, x);
+    cout << (x << 32 | x) << ' ' << x << '\n';
+}
+
+int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie, cout.tie(NULL);
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            solve();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cc-128be.cc

@@ -0,0 +1,325 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (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) decltype(container)::value_type
+
+/* type aliases */
+using ll = int64_t;
+using ull = uint64_t;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+
+/* random */
+
+mt19937 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;
+    }
+};
+
+/* build data structures */
+#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 pa(a) __AS_PROCEDURE(__typeof(a) pa; pa.push_back({}); for (auto&&x : a) pa.push_back(pa.back() + x);)
+#define sa(a) __AS_PROCEDURE(__typeof(a) sa(a.size() + 1); {int n = a.size(); for (int i = n - 1; i >= 0; --i) sa[i] = sa[i + 1] + a[i];};)
+#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 Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+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))
+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(type, ...) __AS_PROCEDURE(type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (int i = 0; i < (n); ++i) cin >> a[i];)
+#define putvec(a) __AS_PROCEDURE(for (auto&& x : a) cout << x << ' '; cout << endl;)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    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;
+}
+
+/* pops */
+#define poptop(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.top(); q.pop();)
+#define popback(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.back(); q.pop_back();)
+#define popfront(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.front();q.pop_front();)
+
+/* math */
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+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) {
+    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;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 512
+
+void dump() {}
+
+void prep() {}
+
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    void operator+=(const MLL& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLL& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLL& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLL& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLL& rhs) { val = (*this % rhs).val; }
+};
+
+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;
+}
+
+void solve() {
+    using ll = MLL<MDL>;
+    read(int, n, m, k);
+    vector<ll> fact;
+    fact.push_back(1);
+    for (int i = 1; i <= n; ++i) {
+        fact.push_back(fact.back() * i);
+    }
+    // (M - K + 1) powers
+    vector<ll> pw1;
+    pw1.push_back(1);
+    for (int i = 1; i < k; ++i) {
+        pw1.push_back(pw1.back() * (m - k + 1));
+    }
+    // (K - 1) powers
+    vector<ll> pw2;
+    for (int i = 1; i < k; ++i)
+}
+
+int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie, cout.tie(NULL);
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            solve();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1486e.cc

@@ -0,0 +1,320 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (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) decltype(container)::value_type
+
+/* type aliases */
+using ll = int64_t;
+using ull = uint64_t;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+
+/* random */
+
+mt19937 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;
+    }
+};
+
+/* build data structures */
+#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 pa(a) __AS_PROCEDURE(__typeof(a) pa; pa.push_back({}); for (auto&&x : a) pa.push_back(pa.back() + x);)
+#define sa(a) __AS_PROCEDURE(__typeof(a) sa(a.size() + 1); {int n = a.size(); for (int i = n - 1; i >= 0; --i) sa[i] = sa[i + 1] + a[i];};)
+#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 Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+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))
+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(type, ...) __AS_PROCEDURE(type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (int i = 0; i < (n); ++i) cin >> a[i];)
+#define putvec(a) __AS_PROCEDURE(for (auto&& x : a) cout << x << ' '; cout << endl;)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    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;
+}
+
+/* pops */
+#define poptop(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.top(); q.pop();)
+#define popback(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.back(); q.pop_back();)
+#define popfront(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.front();q.pop_front();)
+
+/* math */
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+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) {
+    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;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 512
+
+void dump() {}
+
+void prep() {}
+
+void solve() {
+    read(int, n, m);
+    vector<vector<pii>> ch1(n + 1);
+    while (m--) {
+        read(int, u, v, w);
+        ch1[u].emplace_back(v, w);
+        ch1[v].emplace_back(u, w);
+    }
+    vector<vector<pii>> ch(n + 1);
+    for (int v = 1; v <= n; ++v) {
+        for (auto&& [i, w1] : ch1[v]) {
+            for (auto&& [u, w2] : ch1[i]) {
+                if (u == v) continue;
+                ch[v].emplace_back(u, (w1 + w2) * (w1 + w2));
+                ch[u].emplace_back(v, (w1 + w2) * (w1 + w2));
+            }
+        }
+    }
+    vector<bool> vis(n + 1);
+    vector<ll> dis(n + 1, INFLL);
+    priority_queue<pli, vector<pli>, greater<>> pq;
+    dis[1] = 0;
+    pq.emplace(0, 1);
+    while (pq.size()) {
+        poptop(pq, d, v);
+        continue_or(vis[v], 1);
+        // debug(v);
+        for (auto&& [u, w] : ch[v]) {
+            if (d + w < dis[u]) {
+                dis[u] = d + w;
+                pq.emplace(dis[u], u);
+            }
+        }
+    }
+    for (int i = 1; i <= n; ++i) {
+        cout << (dis[i] == INFLL ? -1 : dis[i]) << " \n"[i == n];
+    }
+}
+
+int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie, cout.tie(NULL);
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            solve();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1526d.cc

@@ -0,0 +1,338 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (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) decltype(container)::value_type
+
+/* type aliases */
+using ll = int64_t;
+using ull = uint64_t;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+
+/* random */
+
+mt19937 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;
+    }
+};
+
+/* build data structures */
+#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 pa(a) __AS_PROCEDURE(__typeof(a) pa; pa.push_back({}); for (auto&&x : a) pa.push_back(pa.back() + x);)
+#define sa(a) __AS_PROCEDURE(__typeof(a) sa(a.size() + 1); {int n = a.size(); for (int i = n - 1; i >= 0; --i) sa[i] = sa[i + 1] + a[i];};)
+#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 Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+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))
+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(type, ...) __AS_PROCEDURE(type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (int i = 0; i < (n); ++i) cin >> a[i];)
+#define putvec(a) __AS_PROCEDURE(for (auto&& x : a) cout << x << ' '; cout << endl;)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    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;
+}
+
+/* pops */
+#define poptop(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.top(); q.pop();)
+#define popback(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.back(); q.pop_back();)
+#define popfront(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.front();q.pop_front();)
+
+/* math */
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+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) {
+    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;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 512
+
+void dump() {}
+
+void prep() {}
+
+constexpr int charset[4] = {'A', 'N', 'O', 'T'};
+
+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;
+    }
+};
+
+void solve() {
+    read(string, a);
+    int n = a.size();
+    unordered_map<char, deque<int>, safe_hash> oc;
+    for (int i = 0; i < n; ++i) {
+        oc[a[i]].push_back(i);;
+    }
+    string res;
+    BIT<int> occ(n);
+    for (int i = 0; i < n; ++i) {
+        char mx;
+        int mx_v = -1;
+        for (auto&& c : charset) {
+            if (oc[c].size() && oc[c].front() - occ.getsum(oc[c].front()) > mx_v) {
+                mx_v = oc[c].front() - occ.getsum(oc[c].front());
+                mx = c;
+            }
+        }
+        assert(mx_v != -1);
+        debug(i), debug(mx), debug(mx_v);
+        cerr << "oc = ";
+        for (auto&& [u, v] : oc) {
+            cerr << "{" << u << ", ";
+            for (auto&& x : v) cerr << x << ' ';
+            cerr << "}, ";
+        }
+        cerr << endl;
+        res.push_back(mx);
+        oc[mx].pop_front();
+        occ.add(oc[mx].front(), 1);
+    }
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie, cout.tie(NULL);
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            solve();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1530e.cc

@@ -0,0 +1,410 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (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) decltype(container)::value_type
+
+/* type aliases */
+using ll = int64_t;
+using ull = uint64_t;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+
+/* random */
+
+mt19937 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;
+    }
+};
+
+/* build data structures */
+#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 pa(a) __AS_PROCEDURE(__typeof(a) pa; pa.push_back({}); for (auto&&x : a) pa.push_back(pa.back() + x);)
+#define sa(a) __AS_PROCEDURE(__typeof(a) sa(a.size() + 1); {int n = a.size(); for (int i = n - 1; i >= 0; --i) sa[i] = sa[i + 1] + a[i];};)
+#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 Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+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))
+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(type, ...) __AS_PROCEDURE(type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (int i = 0; i < (n); ++i) cin >> a[i];)
+#define putvec(a) __AS_PROCEDURE(for (auto&& x : a) cout << x << ' '; cout << endl;)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    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;
+}
+
+/* pops */
+#define poptop(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.top(); q.pop();)
+#define popback(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.back(); q.pop_back();)
+#define popfront(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.front();q.pop_front();)
+
+/* math */
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+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) {
+    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;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 512
+
+void dump() {}
+
+void prep() {}
+
+void solve() {
+    read(string, a);
+    int n = a.size();
+    array<int, 26> slot {};
+    for (auto&& x : a) slot[x - 97] += 1;
+    int uq = 26 - count(slot.begin(), slot.end(), 0);
+    auto check = [&] (int k) -> optional<string> {
+        // only if uq >= 3
+        cerr << "Checking k = " << k << endl;
+        for (int i = 0; i < 26; ++i) {
+            if (!slot[i]) continue;
+            debug(i);
+            array<int, 26> slot1 = slot;
+            string res;
+            char mark = -1;
+            while (slot1[i]) {
+                int use = min(slot1[i], k + 1);
+                slot1[i] -= use;
+                res += string(use, char(i + 97));
+                if (res.size() == n) break;
+                int f = 0;
+                for (int j = 0; j < 26; ++j) {
+                    if (slot1[j] && j != mark && j != i) {
+                        res.push_back(j + 97);
+                        --slot1[j];
+                        if (mark == -1) {
+                            mark = j;
+                        }
+                        f = 1;
+                        break;
+                    }
+                }
+                if (f == 0) {
+                    goto fi;
+                }
+            }
+            for (int j = 0; j < 26; ++j) {
+                if (slot1[j]) res += string(slot1[j], char(j + 97));
+            }
+            assert(res.size() == n);
+            return res;
+        fi:
+            ;;
+        }
+        return nullopt;
+    };
+    char x = -1, y = -1;
+    auto check2 = [&] (int k, char x, char y) -> optional<string> {
+        for (int i = 0; i < 2; ++i) {
+            string res;
+            int cb = 0;
+            array<int, 26> slot1 = slot;
+            if (slot[x] < k + 1) {
+                res = string(slot[x], x + 97) + string(slot[y], y + 97);
+                return res;
+            } else {
+                res = string(k + 1, x + 97);
+                slot1[x] -= k + 1;
+                cb = k;
+                int flag = 1;
+                for (int i = k + 1; i < n; ++i) {
+                    if (cb == k) {
+                        if (!slot1[y]) {
+                            flag = 0;
+                            break;
+                        } else {
+                            --slot1[y];
+                            res.push_back(y + 97);
+                            cb = 0;
+                        }
+                    } else {
+                        if (!slot1[x]) {
+                            assert(slot1[y]);
+                            --slot1[y];
+                            res.push_back(y + 97);
+                            cb = 0;
+                        } else {
+                            --slot1[x];
+                            res.push_back(x + 97);
+                            cb += 1;
+                        }
+                    }
+                }
+                if (flag) {
+                    assert(res.size() == n);
+                    return res;
+                }
+            }
+            swap(x, y);
+        }
+        return nullopt;
+    };
+    if (uq == 1) {
+        cout << a << '\n';
+        return;
+    } else if (uq == 2) {
+        for (int i = 0; i < 26; ++i) {
+            if (!slot[i]) continue;
+            if (x != -1) {
+                y = i;
+                break;
+            } else {
+                x = i;
+            }
+        }
+        int l = 0, r = n;
+        while (l < r) {
+            int mid = l + r >> 1;
+            if (check2(mid, x, y) != nullopt) {
+                r = mid;
+            } else {
+                l = mid + 1;
+            }
+        }
+        cout << check2(l, x, y).value() << '\n';
+        return;
+    }
+    int l = 0, r = n;
+    while (l < r) {
+        int mid = l + r >> 1;
+        if (check(mid) != nullopt) {
+            r = mid; 
+        } else {
+            l = mid + 1;
+        }
+    }
+    cout << check(l).value() << '\n';
+}
+
+int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie, cout.tie(NULL);
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            solve();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/codefun-1792.cc

@@ -0,0 +1,300 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (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) decltype(container)::value_type
+
+/* type aliases */
+using ll = int64_t;
+using ull = uint64_t;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+
+/* random */
+
+mt19937 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;
+    }
+};
+
+/* build data structures */
+#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 pa(a) __AS_PROCEDURE(__typeof(a) pa; pa.push_back({}); for (auto&&x : a) pa.push_back(pa.back() + x);)
+#define sa(a) __AS_PROCEDURE(__typeof(a) sa(a.size() + 1); {int n = a.size(); for (int i = n - 1; i >= 0; --i) sa[i] = sa[i + 1] + a[i];};)
+#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 Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+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))
+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(type, ...) __AS_PROCEDURE(type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (int i = 0; i < (n); ++i) cin >> a[i];)
+#define putvec(a) __AS_PROCEDURE(for (auto&& x : a) cout << x << ' '; cout << endl;)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    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;
+}
+
+/* pops */
+#define poptop(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.top(); q.pop();)
+#define popback(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.back(); q.pop_back();)
+#define popfront(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.front();q.pop_front();)
+
+/* math */
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+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) {
+    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;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 512
+
+void dump() {}
+
+void prep() {}
+
+void solve() {
+    read(int, n);
+    priority_queue<ll> pq;
+    while (n--) {
+        read(int, x);
+        pq.emplace(x);
+    }
+    ll res = 0;
+    while (pq.size()) {
+        ll x = pq.top();  pq.pop();
+        if (x == 0) continue;
+        int v = 63 - __builtin_clzll(x);
+        res |= 1LL << v;
+        x ^= 1LL << v;
+        pq.push(x);
+    }
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie, cout.tie(NULL);
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            solve();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/codefun-1793.cc

@@ -0,0 +1,291 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (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) decltype(container)::value_type
+
+/* type aliases */
+using ll = int64_t;
+using ull = uint64_t;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+
+/* random */
+
+mt19937 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;
+    }
+};
+
+/* build data structures */
+#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 pa(a) __AS_PROCEDURE(__typeof(a) pa; pa.push_back({}); for (auto&&x : a) pa.push_back(pa.back() + x);)
+#define sa(a) __AS_PROCEDURE(__typeof(a) sa(a.size() + 1); {int n = a.size(); for (int i = n - 1; i >= 0; --i) sa[i] = sa[i + 1] + a[i];};)
+#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 Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+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))
+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(type, ...) __AS_PROCEDURE(type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (int i = 0; i < (n); ++i) cin >> a[i];)
+#define putvec(a) __AS_PROCEDURE(for (auto&& x : a) cout << x << ' '; cout << endl;)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    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;
+}
+
+/* pops */
+#define poptop(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.top(); q.pop();)
+#define popback(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.back(); q.pop_back();)
+#define popfront(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.front();q.pop_front();)
+
+/* math */
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+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) {
+    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;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 512
+
+void dump() {}
+
+void prep() {}
+
+void solve() {
+    read(int, n, k);
+    unordered_set<int, safe_hash> st;
+    while (n--) {
+        read(int, x);
+        st.emplace(x);
+    }
+    cout << min(ull(k), st.size()) << '\n';
+}
+
+int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie, cout.tie(NULL);
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            solve();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/codefun-1794.cc

@@ -0,0 +1,296 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (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) decltype(container)::value_type
+
+/* type aliases */
+using ll = int64_t;
+using ull = uint64_t;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+
+/* random */
+
+mt19937 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;
+    }
+};
+
+/* build data structures */
+#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 pa(a) __AS_PROCEDURE(__typeof(a) pa; pa.push_back({}); for (auto&&x : a) pa.push_back(pa.back() + x);)
+#define sa(a) __AS_PROCEDURE(__typeof(a) sa(a.size() + 1); {int n = a.size(); for (int i = n - 1; i >= 0; --i) sa[i] = sa[i + 1] + a[i];};)
+#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 Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+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))
+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(type, ...) __AS_PROCEDURE(type __VA_ARGS__; __read(__VA_ARGS__);)
+#define readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (int i = 0; i < (n); ++i) cin >> a[i];)
+#define putvec(a) __AS_PROCEDURE(for (auto&& x : a) cout << x << ' '; cout << endl;)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    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;
+}
+
+/* pops */
+#define poptop(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.top(); q.pop();)
+#define popback(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.back(); q.pop_back();)
+#define popfront(q, ...) __AS_PROCEDURE(auto [__VA_ARGS__] = q.front();q.pop_front();)
+
+/* math */
+constexpr inline int lg2(ll x) { return x == 0 ? -1 : sizeof(ll) * 8 - 1 - __builtin_clzll(x); }
+
+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) {
+    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;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 512
+
+void dump() {}
+
+void prep() {}
+
+void solve() {
+    read(int, n);
+    readvec(ll, a, n);
+    vector<ll> res;
+    int next = 0;
+    for (int i = 0; i < n - 1; i = next) {
+        next = i + 1;
+        while (next < n - 1 && a[next] > a[i]) ++next;
+        res.push_back(a[i] * (next - i));
+        for (int j = 0; j < next - i - 1; ++j) res.push_back(0);
+    }
+    res.push_back(0);
+    putvec(res);
+}
+
+int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie, cout.tie(NULL);
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            solve();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/d.cc

@@ -1,4 +1,3 @@
-#pragma GCC optimize("Ofast")
 /////////////////////////////////////////////////////////
 /**
  * Useful Macros
@@ -182,6 +181,22 @@ ll inverse(ll a, ll b) {
     return mod(x, b);
 }
 
+vector<tuple<int, int, ll>> decompose(ll x) {
+    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;
+}
+
 /* string algorithms */
 vector<int> calc_next(string t) {  // pi function of t
   int n = (int)t.length();
@@ -233,38 +248,90 @@ int period(string s) {  // find the length of shortest recurring period
 }
 /////////////////////////////////////////////////////////
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 512
 
 void dump() {}
 
 void prep() {}
 
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    void operator+=(const MLL& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLL& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLL& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLL& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLL& rhs) { val = (*this % rhs).val; }
+};
+
+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;
+}
+
 void solve() {
-    read(int, n, m, x);
+    using ll = MLL<PRIME>;
-    vector<vector<bool>> dp(m + 1, vector<bool>(n));
+    read(int, n);
-    dp[0][x - 1] = 1;
+    vector<vector<pii>> ch(n + 1);
-    for (int i = 1; i <= m; ++i) {
+    for (int i = 0; i < n - 1; ++i) {
-        read(int, r), read(char, op);
+        read(int, u, v);
-        if (op != '1') {
+        ch[u].emplace_back(v, i);
-            for (int j = 0; j < n; ++j) {
+        ch[v].emplace_back(u, i);
-                if (dp[i - 1][j]) dp[i][mod(j + r, n)] = 1;
+    }
+    read(ll, e);
+    int root = -1;
+    int root_edge;
+    for (int i = 1; i <= n; ++i) {
+        if (ch[i].size() == 1) {
+            root = i;
+            root_edge = ch[i][0].second;
+            break;
         }
     }
-        if (op != '0') {
+    if (root == -1) {
-            for (int j = 0; j < n; ++j) {
+        cout << -1 << '\n';
-                if (dp[i - 1][j]) dp[i][mod(j - r, n)] = 1;
+        return;
+    }
+    vector<int> dis(n + 1);
+    auto dfs = [&] (auto dfs, int v, int pa, int d) -> void {
+        dis[v] = d;
+        for (auto&& [u, _] : ch[v]) {
+            if (u == pa) continue;
+            dfs(dfs, u, v, d + 1);
+        }
+    };
+    dfs(dfs, root, 0, 0);
+    ll res = accumulate(dis.begin(), dis.end(), ll(0)) + *max_element(dis.begin(), dis.end());
+    ll target = e * n;
+    ll tar = (target - res) / n + 1;
+    int64_t tar_val = tar.val;
+    if (tar_val <= n + 2) {
+        tar_val += PRIME;
+    }
+    for (int i = 0; i < n - 1; ++i) {
+        if (i == root_edge) {
+        cout << tar_val << '\n';
+        } else {
+            cout << 1 << '\n';
         }
     }
 }
-    cout << count(dp[m].begin(), dp[m].end(), 1) << "\n";
-    for (int i = 0; i < n; ++i) {
-        if (dp[m][i]) cout << i + 1 << ' ';
-    }
-    cout << "\n";
-}
 
 int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
     untie, cout.tie(NULL);
     prep();
 #ifdef SINGLE_TEST_CASE

src/bin/e.cc

@@ -182,6 +182,22 @@ ll inverse(ll a, ll b) {
     return mod(x, b);
 }
 
+vector<tuple<int, int, ll>> decompose(ll x) {
+    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;
+}
+
 /* string algorithms */
 vector<int> calc_next(string t) {  // pi function of t
   int n = (int)t.length();
@@ -233,24 +249,61 @@ int period(string s) {  // find the length of shortest recurring period
 }
 /////////////////////////////////////////////////////////
 
-#define SINGLE_TEST_CASE
+// #define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 512
 
 void dump() {}
 
 void prep() {}
 
-void solve() {
+template <ll mdl> struct MLL {
-    read(int, n);
+    ll val;
-    read(string, a);
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
-    for (int i = 0; i < n; ++i) {
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
-        for (int j = i + 1; j < n; ++j) {
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    void operator+=(const MLL& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLL& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLL& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLL& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLL& rhs) { val = (*this % rhs).val; }
+};
 
+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;
 }
+
+void solve() {
+    using ll = MLL<MDL>;
+    read(int, n, m, k);
+    vector<ll> fact;
+    fact.push_back(1);
+    for (int i = 1; i <= n; ++i) {
+        fact.push_back(fact.back() * i);
+    }
+    // (M - K + 1) powers
+    vector<ll> pw1;
+    pw1.push_back(1);
+    for (int i = 1; i < k; ++i) {
+        pw1.push_back(pw1.back() * (m - k + 1));
+    }
+    // (K - 1) powers
+    vector<ll> pw2;
+    for (int i = 1; i < k; ++i)
 }
 
 int main() {
+#if __cplusplus < 201703L || defined(_MSC_VER) && !defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
     untie, cout.tie(NULL);
     prep();
 #ifdef SINGLE_TEST_CASE

src/bin/std.in

@@ -1,13 +1,4 @@
-6 7
-2 1
-2 3
-3 5
-1 4
-6 1
-5 6
-4 5
 3
-1 4
+1 2
-5 1
+2 3
-4 5
+665496238
-

src/bin/template.cc

@@ -249,7 +249,7 @@ int period(string s) {  // find the length of shortest recurring period
 }
 /////////////////////////////////////////////////////////
 
-#define SINGLE_TEST_CASE
+// #define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 512
 
 void dump() {}
@@ -257,6 +257,7 @@ void dump() {}
 void prep() {}
 
 void solve() {
+    
 }
 
 int main() {

src/bin/test.py

@@ -1,4 +1,10 @@
-from random import randint
+def diff(a, b):
-print(500000)
+    n = len(a)
-for i in range(500000):
+    res = 0
-    print(randint(7800000, 7900000))
+    for i in range(n):
+        if a[i] != b[i]:
+            for j in range(i + 1, n):
+                if b[j] == a[i]:
+                    res += j - i
+                    b = b[:i] + b[j] + b[i+1:j] + b[j+1:]
+    return res

src/bin/test1.py

@@ -0,0 +1,11 @@
+def diff(a, b):
+    n = len(a)
+    res = 0
+    for i in range(n):
+        if a[i] != b[i]:
+            for j in range(i + 1, n):
+                if b[j] == a[i]:
+                    res += j - i
+                    b = b[:i] + b[j] + b[i:j] + b[j+1:]
+                    break
+    return res