backup

src/bin/b.cc

@@ -15,7 +15,7 @@ using namespace std;
 #define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
 constexpr void __() {}
 #define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
-#define __as_typeof(container) decltype(container)::value_type
+#define __as_typeof(container) remove_reference<decltype(container)>::type
 
 /* type aliases */
 #if LONG_LONG_MAX != INT64_MAX
@@ -67,6 +67,8 @@ using tiid = tuple<int, int, ld>;
 using tiii = tuple<int, int, int>;
 template <typename T> using max_heap = priority_queue<T>;
 template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
 
 /* constants */
 constexpr int INF = 0x3f3f3f3f;
@@ -201,10 +203,17 @@ template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, s
 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 readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
 #define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
 #define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
 template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
     out << "{" << p.first << ", " << p.second << "}";
     return out;
@@ -372,17 +381,42 @@ template <ll mdl> struct MLL {
     void operator/=(const MLL& rhs) { val = (*this / rhs).val; }
     void operator%=(const MLL& rhs) { val = (*this % rhs).val; }
 };
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
 
 template <ll mdl>
 ostream& operator<<(ostream& out, const MLL<mdl>& num) {
     return out << num.val;
 }
 
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
 template <ll mdl>
 istream& operator>>(istream& in, MLL<mdl>& num) {
     return in >> num.val;
 }
 
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
 // miscancellous
 template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
     std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
@@ -390,32 +424,73 @@ template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, Ran
 template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
     std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
 }
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
 /////////////////////////////////////////////////////////
 
-#define SINGLE_TEST_CASE
+// #define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
 
 void dump() {}
 
 void dump_ignore() {}
 
-void prep() {}
+void prep() {
+}
 
 void solve() {
-    read(int, n, k);
+    read(int, n);
     readvec(int, a, n);
-    int ptr = 0;
-    int empty = k;
-    int cnt = 0;
-    while (ptr < n) {
-        if (empty < a[ptr]) {
-            cnt += 1;
-            empty = k;
-        } else {
-            empty -= a[ptr++];
+    int mn_l = 1;
+    for (int b = 0; b < 20; ++b) {
+        int max_dist = 0;
+        int prev_1 = -1;
+        for (int i = 0; i < n; ++i) {
+            if (a[i] >> b & 1) {
+                max_dist = max(max_dist, i - prev_1);
+                prev_1 = i;
             }
         }
-    cout << cnt + 1 << '\n';
+        if (prev_1 != -1) {
+            max_dist = max(max_dist, n - prev_1);
+            mn_l = max(mn_l, max_dist);
+        }
+    }
+    cout << mn_l << '\n';
 }
 
 int main() {
@@ -430,7 +505,7 @@ int main() {
     read(int, t);
     for (int i = 0; i < t; ++i) {
 #ifdef DUMP_TEST_CASE
-        if (t < (DUMP_TEST_CASE)) {
+        if (t != (TOT_TEST_CASE)) {
             solve();
         } else if (i + 1 == (DUMP_TEST_CASE)) {
             dump();

src/bin/c.cc

@@ -15,7 +15,7 @@ using namespace std;
 #define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
 constexpr void __() {}
 #define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
-#define __as_typeof(container) decltype(container)::value_type
+#define __as_typeof(container) remove_reference<decltype(container)>::type
 
 /* type aliases */
 #if LONG_LONG_MAX != INT64_MAX
@@ -67,6 +67,8 @@ using tiid = tuple<int, int, ld>;
 using tiii = tuple<int, int, int>;
 template <typename T> using max_heap = priority_queue<T>;
 template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
 
 /* constants */
 constexpr int INF = 0x3f3f3f3f;
@@ -201,10 +203,17 @@ template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, s
 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 readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
 #define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
 #define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
 template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
     out << "{" << p.first << ", " << p.second << "}";
     return out;
@@ -372,17 +381,42 @@ template <ll mdl> struct MLL {
     void operator/=(const MLL& rhs) { val = (*this / rhs).val; }
     void operator%=(const MLL& rhs) { val = (*this % rhs).val; }
 };
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
 
 template <ll mdl>
 ostream& operator<<(ostream& out, const MLL<mdl>& num) {
     return out << num.val;
 }
 
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
 template <ll mdl>
 istream& operator>>(istream& in, MLL<mdl>& num) {
     return in >> num.val;
 }
 
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
 // miscancellous
 template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
     std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
@@ -390,56 +424,80 @@ template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, Ran
 template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
     std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
 }
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
 /////////////////////////////////////////////////////////
 
-#define SINGLE_TEST_CASE
+// #define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
 
 void dump() {}
 
 void dump_ignore() {}
 
-void prep() {}
-
-template<typename T>
-struct BIT {
-    int n;
-    vector<T> c;
-    BIT(size_t n) : n(n), c(n + 1) {}
-    void add(size_t i, const T& k) {
-        while (i <= n) {
-            c[i] += k;
-            i += lowbit(i);
-        }
-    }
-    T getsum(size_t i) {
-        T res = {};
-        while (i) {
-            res += c[i];
-            i -= lowbit(i);
-        }
-        return res;
-    }
-};
+void prep() {
+}
 
 void solve() {
     read(int, n);
-    readvec(ll, a, n);
-    auto [N, mp] = discretize<ll>(a.begin(), a.end());
-    BIT<int> tr(N);
-    ll res = 0;
-    ll sum = 0;
-    for (int i = n - 1; ~i; --i) {
-        res += a[i] * (n - 1 - i) + sum;
-        auto it = mp.lower_bound(100'000'000 - a[i]);
-        if (it != mp.end()) {
-            ll cnt = tr.getsum(N) - tr.getsum(max((unsigned long)0, it->second - 1));
-            res -= cnt * 100'000'000;
+    readvec(int, a, n);
+    vector<int> pos;
+    for (int i = 1; i < n; i += 2) {
+        if (a[i] == 1) {
+            i += 1;
         }
-        tr.add(mp[a[i]], 1);
-        sum += a[i];
+        if (i == n) break;
+        pos.emplace_back(i);
     }
-    cout << res << '\n';
+    int target = n;
+    vector<int> res(n);
+    sort_by_key(pos.begin(), pos.end(), [&] (int i) { return a[i]; });
+    for (auto&& i : pos) {
+        res[i] = target--;
+    }
+    vector<int> rem;
+    for (int i = 0; i < n; ++i) {
+        if (not res[i]) rem.emplace_back(i);
+    }
+    sort_by_key(rem.begin(), rem.end(), [&] (int i) { return a[i]; });
+    for (auto&& i : rem) {
+        res[i] = target--;
+    }
+    putvec(res);
 }
 
 int main() {
@@ -454,7 +512,7 @@ int main() {
     read(int, t);
     for (int i = 0; i < t; ++i) {
 #ifdef DUMP_TEST_CASE
-        if (t < (DUMP_TEST_CASE)) {
+        if (t != (TOT_TEST_CASE)) {
             solve();
         } else if (i + 1 == (DUMP_TEST_CASE)) {
             dump();

src/bin/cf-1721d.cc

@@ -0,0 +1,563 @@
+#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) remove_reference<decltype(container)>::type
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+using ld = long double;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+using pid = pair<int, ld>;
+using pdi = pair<ld, int>;
+using pld = pair<ll, ld>;
+using pdl = pair<ld, ll>;
+using pdd = pair<ld, ld>;
+using tlll = tuple<ll, ll, ll>;
+using tlld = tuple<ll, ll, ld>;
+using tlli = tuple<ll, ll, int>;
+using tldl = tuple<ll, ld, ll>;
+using tldd = tuple<ll, ld, ld>;
+using tldi = tuple<ll, ld, int>;
+using tlil = tuple<ll, int, ll>;
+using tlid = tuple<ll, int, ld>;
+using tlii = tuple<ll, int, int>;
+using tdll = tuple<ld, ll, ll>;
+using tdld = tuple<ld, ll, ld>;
+using tdli = tuple<ld, ll, int>;
+using tddl = tuple<ld, ld, ll>;
+using tddd = tuple<ld, ld, ld>;
+using tddi = tuple<ld, ld, int>;
+using tdil = tuple<ld, int, ll>;
+using tdid = tuple<ld, int, ld>;
+using tdii = tuple<ld, int, int>;
+using till = tuple<int, ll, ll>;
+using tild = tuple<int, ll, ld>;
+using tili = tuple<int, ll, int>;
+using tidl = tuple<int, ld, ll>;
+using tidd = tuple<int, ld, ld>;
+using tidi = tuple<int, ld, int>;
+using tiil = tuple<int, int, ll>;
+using tiid = tuple<int, int, ld>;
+using tiii = tuple<int, int, int>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define 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 edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w), ch[v].emplace_back(u, w);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w);)
+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 (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+
+/* 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) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    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; }
+};
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+void solve() {
+    read(int, n);
+    readvec(int, a, n);
+    readvec(int, b, n);
+    auto partition = [&] (int l, int r, int bit) -> void {
+        int j = l;
+        for (int i = l; i <= r; ++i) {
+            while (j + 1 <= i and (a[j] >> bit & 1) == 1) {
+                ++j;
+            }
+            if ((a[i] >> bit & 1) == 1 and (a[j] >> bit & 1) == 0) {
+                swap(a[i], a[j]);
+                ++j;
+            }
+        }
+        j = l;
+        for (int i = l; i <= r; ++i) {
+            while (j + 1 <= i and (b[j] >> bit & 1) == 0) {
+                ++j;
+            }
+            if ((b[i] >> bit & 1) == 0 and (b[j] >> bit & 1) == 1) {
+                swap(b[i], b[j]);
+                ++j;
+            }
+        }
+    };
+    unordered_set<int, safe_hash> bar = {n - 1};
+    for (int bit = 30; ~bit; --bit) {
+        int cnt_a = 0, cnt_b = 0;
+        int f = 1;
+        for (int i = 0; i < n; ++i) {
+            if ((a[i] >> bit & 1) == 1) cnt_a += 1;
+            if ((b[i] >> bit & 1) == 0) cnt_b += 1;
+            if (bar.count(i)) {
+                if (cnt_a != cnt_b) {
+                    f = 0;
+                    break;
+                }
+                cnt_a = 0, cnt_b = 0;
+            }
+        }
+        if (f) {
+            int prev = 0;
+            for (int i = 0; i < n; ++i) {
+                if (bar.count(i)) {
+                    partition(prev, i, bit);
+                    for (int j = prev; j <= i; ++j) {
+                        if ((a[j] >> bit & 1) == 0) {
+                            if (j != prev) bar.emplace(j - 1);
+                            break;
+                        }
+                    }
+                    prev = i + 1;
+                }
+            }
+        }
+    }
+    int res = ~0;
+    for (int i = 0; i < n; ++i) {
+        res &= a[i] ^ b[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 (t < (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1729e.cc

@@ -0,0 +1,517 @@
+#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) remove_reference<decltype(container)>::type
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+using ld = long double;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+using pid = pair<int, ld>;
+using pdi = pair<ld, int>;
+using pld = pair<ll, ld>;
+using pdl = pair<ld, ll>;
+using pdd = pair<ld, ld>;
+using tlll = tuple<ll, ll, ll>;
+using tlld = tuple<ll, ll, ld>;
+using tlli = tuple<ll, ll, int>;
+using tldl = tuple<ll, ld, ll>;
+using tldd = tuple<ll, ld, ld>;
+using tldi = tuple<ll, ld, int>;
+using tlil = tuple<ll, int, ll>;
+using tlid = tuple<ll, int, ld>;
+using tlii = tuple<ll, int, int>;
+using tdll = tuple<ld, ll, ll>;
+using tdld = tuple<ld, ll, ld>;
+using tdli = tuple<ld, ll, int>;
+using tddl = tuple<ld, ld, ll>;
+using tddd = tuple<ld, ld, ld>;
+using tddi = tuple<ld, ld, int>;
+using tdil = tuple<ld, int, ll>;
+using tdid = tuple<ld, int, ld>;
+using tdii = tuple<ld, int, int>;
+using till = tuple<int, ll, ll>;
+using tild = tuple<int, ll, ld>;
+using tili = tuple<int, ll, int>;
+using tidl = tuple<int, ld, ll>;
+using tidd = tuple<int, ld, ld>;
+using tidi = tuple<int, ld, int>;
+using tiil = tuple<int, int, ll>;
+using tiid = tuple<int, int, ld>;
+using tiii = tuple<int, int, int>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define 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 edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w), ch[v].emplace_back(u, w);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w);)
+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 (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+
+/* 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) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    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; }
+};
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+void solve() {
+    for (int i = 2; i <= 1e18; ++i) {
+        cout << "? 1 " << i << endl;
+        read(ll, x);
+        cout << "? " << i << " 1" << endl;
+        read(ll, y);
+        if (x != y) {
+            cout << "! " << x + y << endl;
+            return;
+        }
+        if (x == -1) {
+            cout << "! " << i - 1 << endl;
+            return;
+        }
+    }
+    exit(825);
+}
+
+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 (t < (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1734d.cc

@@ -0,0 +1,550 @@
+#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) remove_reference<decltype(container)>::type
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+using ld = long double;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+using pid = pair<int, ld>;
+using pdi = pair<ld, int>;
+using pld = pair<ll, ld>;
+using pdl = pair<ld, ll>;
+using pdd = pair<ld, ld>;
+using tlll = tuple<ll, ll, ll>;
+using tlld = tuple<ll, ll, ld>;
+using tlli = tuple<ll, ll, int>;
+using tldl = tuple<ll, ld, ll>;
+using tldd = tuple<ll, ld, ld>;
+using tldi = tuple<ll, ld, int>;
+using tlil = tuple<ll, int, ll>;
+using tlid = tuple<ll, int, ld>;
+using tlii = tuple<ll, int, int>;
+using tdll = tuple<ld, ll, ll>;
+using tdld = tuple<ld, ll, ld>;
+using tdli = tuple<ld, ll, int>;
+using tddl = tuple<ld, ld, ll>;
+using tddd = tuple<ld, ld, ld>;
+using tddi = tuple<ld, ld, int>;
+using tdil = tuple<ld, int, ll>;
+using tdid = tuple<ld, int, ld>;
+using tdii = tuple<ld, int, int>;
+using till = tuple<int, ll, ll>;
+using tild = tuple<int, ll, ld>;
+using tili = tuple<int, ll, int>;
+using tidl = tuple<int, ld, ll>;
+using tidd = tuple<int, ld, ld>;
+using tidi = tuple<int, ld, int>;
+using tiil = tuple<int, int, ll>;
+using tiid = tuple<int, int, ld>;
+using tiii = tuple<int, int, int>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define 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 edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w), ch[v].emplace_back(u, w);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w);)
+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 (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+
+/* 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) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    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; }
+};
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+void solve() {
+    read(int, n, k);
+    --k;
+    readvec(int, a, n);
+    vector<ll> ps(n + 1);
+    for (int i = 1; i <= n; ++i) {
+        ps[i] = ps[i - 1] + a[i - 1];
+    }
+    {
+        int f = 1;
+        int l = k, max_l = k;
+        for (int i = k; i < n; ++i) {
+            if (ps[i + 1] - ps[max_l] < 0) {
+                f = 0;
+                break;
+            }
+            while (l - 1 >= 0 and ps[i + 1] >= ps[l - 1]) {
+                l -= 1;
+                if (ps[l] < ps[max_l]) {
+                    max_l = l;
+                }
+            }
+        }
+        if (f) {
+            cout << "YES\n";
+            return;
+        }
+    }
+    {
+        int f = 1;
+        int r = k + 1, max_r = k + 1;
+        for (int i = k; ~i; --i) {
+            if (ps[max_r] - ps[i] < 0) {
+                f = 0;
+                break;
+            }
+            while (r + 1 <= n and ps[r + 1] >= ps[i]) {
+                r += 1;
+                if (ps[r] >= ps[max_r]) {
+                    max_r = r;
+                }
+            }
+        }
+        if (f) {
+            cout << "YES\n";
+            return;
+        }
+    }
+    cout << "NO\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 (t < (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1738e.cc

@@ -0,0 +1,538 @@
+#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) remove_reference<decltype(container)>::type
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+using ld = long double;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+using pid = pair<int, ld>;
+using pdi = pair<ld, int>;
+using pld = pair<ll, ld>;
+using pdl = pair<ld, ll>;
+using pdd = pair<ld, ld>;
+using tlll = tuple<ll, ll, ll>;
+using tlld = tuple<ll, ll, ld>;
+using tlli = tuple<ll, ll, int>;
+using tldl = tuple<ll, ld, ll>;
+using tldd = tuple<ll, ld, ld>;
+using tldi = tuple<ll, ld, int>;
+using tlil = tuple<ll, int, ll>;
+using tlid = tuple<ll, int, ld>;
+using tlii = tuple<ll, int, int>;
+using tdll = tuple<ld, ll, ll>;
+using tdld = tuple<ld, ll, ld>;
+using tdli = tuple<ld, ll, int>;
+using tddl = tuple<ld, ld, ll>;
+using tddd = tuple<ld, ld, ld>;
+using tddi = tuple<ld, ld, int>;
+using tdil = tuple<ld, int, ll>;
+using tdid = tuple<ld, int, ld>;
+using tdii = tuple<ld, int, int>;
+using till = tuple<int, ll, ll>;
+using tild = tuple<int, ll, ld>;
+using tili = tuple<int, ll, int>;
+using tidl = tuple<int, ld, ll>;
+using tidd = tuple<int, ld, ld>;
+using tidi = tuple<int, ld, int>;
+using tiil = tuple<int, int, ll>;
+using tiid = tuple<int, int, ld>;
+using tiii = tuple<int, int, int>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define 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 edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w), ch[v].emplace_back(u, w);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w);)
+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 (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+
+/* 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) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    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; }
+};
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+void solve() {
+    read(int, n);
+    readvec(int, a, n);
+    int r = n - 1;  // next r
+    ll sum_l = 0, sum_r = 0;
+    vector<tlii> info;
+    int zero_cnt_l = 0, zero_cnt_r = 0;
+    for (int i = 0; i < n; ++i) {
+        if (a[i] == 0) {
+            if (zero_cnt_l == 0) {
+                while (a[r] == 0) {
+                    zero_cnt_r += 1;
+                    --r;
+                }
+            } else {
+                zero_cnt_l += 1;
+            }
+        } else {
+            if (zero_cnt_l > 0) {
+                info.emplace_back(0, zero_cnt_l, zero_cnt_r);
+                zero_cnt_l = 0;
+                zero_cnt_r = 0;
+            }
+            while (sum_r < sum_l) {
+                sum_r += a[r];
+                --r;
+            }
+            if (sum_l == sum_r) {
+                info.emplace_back(sum_l, 0, 0);
+            }
+        }
+    }
+    if (zero_cnt_l > 0) {
+        info.emplace_back(0, zero_cnt_l, zero_cnt_r);
+        zero_cnt_l = 0;
+        zero_cnt_r = 0;
+    }
+}
+
+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 (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1761d.cc

@@ -0,0 +1,529 @@
+#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) remove_reference<decltype(container)>::type
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+using ld = long double;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+using pid = pair<int, ld>;
+using pdi = pair<ld, int>;
+using pld = pair<ll, ld>;
+using pdl = pair<ld, ll>;
+using pdd = pair<ld, ld>;
+using tlll = tuple<ll, ll, ll>;
+using tlld = tuple<ll, ll, ld>;
+using tlli = tuple<ll, ll, int>;
+using tldl = tuple<ll, ld, ll>;
+using tldd = tuple<ll, ld, ld>;
+using tldi = tuple<ll, ld, int>;
+using tlil = tuple<ll, int, ll>;
+using tlid = tuple<ll, int, ld>;
+using tlii = tuple<ll, int, int>;
+using tdll = tuple<ld, ll, ll>;
+using tdld = tuple<ld, ll, ld>;
+using tdli = tuple<ld, ll, int>;
+using tddl = tuple<ld, ld, ll>;
+using tddd = tuple<ld, ld, ld>;
+using tddi = tuple<ld, ld, int>;
+using tdil = tuple<ld, int, ll>;
+using tdid = tuple<ld, int, ld>;
+using tdii = tuple<ld, int, int>;
+using till = tuple<int, ll, ll>;
+using tild = tuple<int, ll, ld>;
+using tili = tuple<int, ll, int>;
+using tidl = tuple<int, ld, ll>;
+using tidd = tuple<int, ld, ld>;
+using tidi = tuple<int, ld, int>;
+using tiil = tuple<int, int, ll>;
+using tiid = tuple<int, int, ld>;
+using tiii = tuple<int, int, int>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define 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 edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w), ch[v].emplace_back(u, w);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w);)
+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 (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+
+/* 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) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    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; }
+};
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+constexpr int MAXN = 1e6 + 10;
+using mll = MLL<MDL>;
+mll fact[MAXN], pw2[MAXN], pw3[MAXN], s[MAXN];
+void prep() {
+    fact[0] = 1, pw2[0] = 1, pw3[0] = 1, s[0] = 0;
+    for (int i = 1; i < MAXN; ++i) {
+        fact[i] = fact[i - 1] * i;
+        pw3[i] = pw3[i - 1] * 3;
+        pw2[i] = pw2[i - 1] * 2;
+        s[i] = 3 * s[i - 1] + 2 * pw3[i - 1];
+    }
+}
+
+mll comb(int n, int k) {
+    if (n < 0 or k < 0 or n < k) return 0;
+    return fact[n] / fact[k] / fact[n - k];
+}
+
+void solve() {
+    read(int, n, k);
+    if (k == 0) {
+        cout << pw3[n] << '\n';
+    } else {
+        mll res = 0;
+        int t = n - k;
+        for (int j = 0; j <= k; ++j) {
+            if (t - j < 0) continue;
+            res += pw3[t - j] * comb(t, j) / k * comb(k, j) * (k * pw3[k - j] - s[k - j]);
+        }
+        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 (t < (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1761e.cc

@@ -0,0 +1,737 @@
+#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) remove_reference<decltype(container)>::type
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+using ld = long double;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+using pid = pair<int, ld>;
+using pdi = pair<ld, int>;
+using pld = pair<ll, ld>;
+using pdl = pair<ld, ll>;
+using pdd = pair<ld, ld>;
+using tlll = tuple<ll, ll, ll>;
+using tlld = tuple<ll, ll, ld>;
+using tlli = tuple<ll, ll, int>;
+using tldl = tuple<ll, ld, ll>;
+using tldd = tuple<ll, ld, ld>;
+using tldi = tuple<ll, ld, int>;
+using tlil = tuple<ll, int, ll>;
+using tlid = tuple<ll, int, ld>;
+using tlii = tuple<ll, int, int>;
+using tdll = tuple<ld, ll, ll>;
+using tdld = tuple<ld, ll, ld>;
+using tdli = tuple<ld, ll, int>;
+using tddl = tuple<ld, ld, ll>;
+using tddd = tuple<ld, ld, ld>;
+using tddi = tuple<ld, ld, int>;
+using tdil = tuple<ld, int, ll>;
+using tdid = tuple<ld, int, ld>;
+using tdii = tuple<ld, int, int>;
+using till = tuple<int, ll, ll>;
+using tild = tuple<int, ll, ld>;
+using tili = tuple<int, ll, int>;
+using tidl = tuple<int, ld, ll>;
+using tidd = tuple<int, ld, ld>;
+using tidi = tuple<int, ld, int>;
+using tiil = tuple<int, int, ll>;
+using tiid = tuple<int, int, ld>;
+using tiii = tuple<int, int, int>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define 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 edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w), ch[v].emplace_back(u, w);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w);)
+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 (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+
+/* 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) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    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; }
+};
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+namespace tarjan {
+    struct mutex_cond {
+        int v1; bool cond1;
+        int v2; bool cond2;
+        mutex_cond(int v1, bool cond1, int v2, bool cond2) : v1(v1), cond1(cond1), v2(v2), cond2(cond2) {}
+    };
+    struct inclusive_cond {
+        int v1; bool cond1;
+        int v2; bool cond2;
+        inclusive_cond(int v1, bool cond1, int v2, bool cond2) : v1(v1), cond1(cond1), v2(v2), cond2(cond2) {}
+    };
+    // Returns the mapping between vertices and their affiliated sccs.
+    vector<int> scc(const vector<vector<int>>& ch) {
+        int n = ch.size();
+        int cnt = 0, scn = 0;
+        vector<int> dfn(n), low(n), vis(n), st;
+        vector<int> br(n);
+        auto tarjan = [&] (auto tarjan, int v) -> void {
+            dfn[v]=low[v]=++cnt;
+            st.push_back(v);
+            vis[v]=1;
+            for(const auto&u:ch[v])
+                if(!dfn[u]) tarjan(tarjan, u),low[v]=min(low[v],low[u]);
+                else if(vis[u])low[v]=min(low[v],dfn[u]);
+            if(dfn[v]==low[v]){
+                ++scn;
+                int u;
+                do u=st.back(), st.pop_back(),vis[u]=0,br[u]=scn; while(u!=v);
+            }
+        };
+        for (int i = 0; i < n; ++i) {
+            if (!dfn[i]) {
+                tarjan(tarjan, i);
+            }
+        }
+        return br;
+    }
+    // This method can eliminate redundant edges or self-loops
+    vector<vector<int>> build_scc(const vector<vector<int>>& ch) {
+        int n = ch.size();
+        auto br = scc(ch);
+        int cnt = *max_element(br.begin(), br.end());
+        vector<unordered_set<int, safe_hash>> rb(cnt + 1);
+        for (int i = 0; i < n; ++i) {
+            for (auto&& u : ch[i]) {
+                if (br[i] != br[u]) rb[br[i]].emplace(br[u]);
+            }
+        }
+        vector<vector<int>> res(cnt + 1);
+        for (int i = 1; i <= cnt; ++i) {
+            res[i] = vector<int>(rb[i].begin(), rb[i].end());
+        }
+        return res;
+    }
+    
+    // This method can eliminate redundant edges or self-loops
+    // return form: (scc size, children of scc)
+    vector<pair<size_t, vector<int>>> build_scc_with_size(const vector<vector<int>>& ch) {
+        int n = ch.size();
+        auto br = scc(ch);
+        int cnt = *max_element(br.begin(), br.end());
+        vector<unordered_set<int, safe_hash>> rb(cnt + 1);
+        for (int i = 0; i < n; ++i) {
+            for (auto&& u : ch[i]) {
+                if (br[i] != br[u]) rb[br[i]].emplace(br[u]);
+            }
+        }
+        vector<pair<size_t, vector<int>>> res(cnt + 1);
+        for (int i = 1; i <= cnt; ++i) {
+            res[i].second = vector<int>(rb[i].begin(), rb[i].end());
+        }
+        for (int i = 1; i <= n; ++i) {
+            res[br[i]].first += 1;
+        }
+        return res;
+    }
+    // indices start from 1, result has `n` items
+    optional<vector<bool>> solve_twosat(int n, const vector<mutex_cond>& conditions) {
+        vector<vector<int>> ch(2 * n + 1);
+        for (auto&& [v1, cond1, v2, cond2] : conditions) {
+            ch[(1 ^ cond1) * n + v1].emplace_back(cond2 * n + v2);
+            ch[(1 ^ cond2) * n + v2].emplace_back(cond1 * n + v1);
+        }
+        auto sccno = scc(ch);
+        for (int i = 1; i <= n; ++i) {
+            if (sccno[i] == sccno[i + n]) {
+                return nullopt;
+            }
+        }
+        vector<bool> res;
+        for (int i = 1; i <= n; ++i) {
+            if (sccno[i] < sccno[i + n]) {
+                res.emplace_back(false);
+            } else {
+                res.emplace_back(true);
+            }
+        }
+        return res;
+    };
+    // indices start from 1, result has `n` items
+    optional<vector<bool>> solve_twosat(int n, const vector<inclusive_cond>& conditions) {
+        vector<mutex_cond> trans_conds;
+        for (auto&& [v1, cond1, v2, cond2] : conditions) {
+            trans_conds.emplace_back(v1, cond1, v2, not cond2);
+        }
+        return solve_twosat(n, trans_conds);
+    }
+
+    // Returns if each vertex is a cut vertex
+    // All indices start from 1
+    vector<int> cut_v(const vector<vector<int>>& ch) {
+        int n = ch.size() - 1;
+        vector<bool> vis(n + 1);
+        vector<int> low(n + 1), dfn(n + 1), flag(n + 1);
+        int cnt = 0;
+        auto dfs = [&] (auto dfs, int v, int pa) -> void {
+            vis[v] = 1;
+            low[v] = dfn[v] = ++cnt;
+            int child = 0;
+            for (auto&& u : ch[v]) {
+                if (not vis[u]) {
+                    ++child;
+                    dfs(dfs, u, v);
+                    low[v] = min(low[v], low[u]);
+                    if (pa != v and low[u] >= dfn[v] and not flag[v]) {
+                        flag[v] = 1;
+                    }
+                } else if (u != pa) {
+                    low[v] = min(low[v], dfn[u]);
+                }
+            }
+            if (pa == v and child >= 2 and not flag[v]) {
+                flag[v] = 1;
+            }
+        };
+        for (int i = 1; i <= n; ++i) {
+            if (not dfn[i]) {
+                dfs(dfs, i, 0);
+            }
+        }
+        return flag;
+    }
+}
+
+void solve() {
+    read(int, n);
+    vector<vector<int>> a(n + 1, vector<int>(n + 1));
+    adj(ch, n);
+    for (int i = 1; i <= n; ++i) {
+        for (int j = 1; j <= n; ++j) {
+            read(char, x);
+            if (x == '1') {
+                a[i][j] = 1;
+                edge(ch, i, j);
+            }
+        }
+    }
+    vector<bool> vis(n + 1);
+    vector<int> no(n + 1);
+    int curr_no = 0;
+    auto dfs = [&] (auto dfs, int v) -> void {
+        if (vis[v]) return;
+        vis[v] = 1;
+        no[v] = curr_no;
+        for (auto&& u : ch[v]) {
+            dfs(dfs, u);
+        }
+    };
+    for (int i = 1; i <= n; ++i) {
+        if (not vis[i]) {
+            ++curr_no;
+            dfs(dfs, i);
+        }
+    }
+    auto cut = tarjan::cut_v(ch);
+    if (count(no.begin() + 1, no.end(), 1) == n) {
+        cout << 0 << '\n';
+    } else {
+        for (int i = 1; i <= curr_no; ++i) {
+            vector<int> sub;
+            for (int j = 1; j <= n; ++j) {
+                if (no[j] == i) {
+                    sub.emplace_back(j);
+                }
+            }
+            int m = sub.size();
+            for (int j = 0; j < m; ++j) {
+                for (int k = j + 1; k < m; ++k) {
+                    if (not a[sub[j]][sub[k]]) {
+                        if (not cut[sub[j]]) {
+                            cout << 1 << '\n';
+                            cout << sub[j] << '\n';
+                            return;
+                        }
+                        if (not cut[sub[k]]) {
+                            cout << 1 << '\n';
+                            cout << sub[k] << '\n';
+                            return;
+                        }
+                    }
+                }
+            }
+            if (sub.size() == 1) {
+                cout << 1 << '\n';
+                cout << sub[0] << '\n';
+                return;
+            }
+        }
+        if (curr_no >= 3) {
+            cout << 2 << '\n';
+            cout << 1 << ' ';
+            for (int i = 1; i <= n; ++i) {
+                if (no[i] != no[1]) {
+                    cout << i << '\n';
+                    return;
+                }
+            }
+            assert(false);
+        } else {
+            vector<int> one, two;
+            for (int i = 1; i <= n; ++i) {
+                if (no[i] == 1) {
+                    one.emplace_back(i);
+                } else {
+                    two.emplace_back(i);
+                }
+            }
+            if (one.size() < two.size()) {
+                cout << one.size() << '\n';
+                putvec(one);
+            } else {
+                cout << two.size() << '\n';
+                putvec(two);
+            }
+        }
+    }
+}
+
+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 (t < (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1766d.cc

@@ -0,0 +1,518 @@
+#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) remove_reference<decltype(container)>::type
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+using ld = long double;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+using pid = pair<int, ld>;
+using pdi = pair<ld, int>;
+using pld = pair<ll, ld>;
+using pdl = pair<ld, ll>;
+using pdd = pair<ld, ld>;
+using tlll = tuple<ll, ll, ll>;
+using tlld = tuple<ll, ll, ld>;
+using tlli = tuple<ll, ll, int>;
+using tldl = tuple<ll, ld, ll>;
+using tldd = tuple<ll, ld, ld>;
+using tldi = tuple<ll, ld, int>;
+using tlil = tuple<ll, int, ll>;
+using tlid = tuple<ll, int, ld>;
+using tlii = tuple<ll, int, int>;
+using tdll = tuple<ld, ll, ll>;
+using tdld = tuple<ld, ll, ld>;
+using tdli = tuple<ld, ll, int>;
+using tddl = tuple<ld, ld, ll>;
+using tddd = tuple<ld, ld, ld>;
+using tddi = tuple<ld, ld, int>;
+using tdil = tuple<ld, int, ll>;
+using tdid = tuple<ld, int, ld>;
+using tdii = tuple<ld, int, int>;
+using till = tuple<int, ll, ll>;
+using tild = tuple<int, ll, ld>;
+using tili = tuple<int, ll, int>;
+using tidl = tuple<int, ld, ll>;
+using tidd = tuple<int, ld, ld>;
+using tidi = tuple<int, ld, int>;
+using tiil = tuple<int, int, ll>;
+using tiid = tuple<int, int, ld>;
+using tiii = tuple<int, int, int>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define 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 edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w), ch[v].emplace_back(u, w);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w);)
+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 (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+
+/* 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) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    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; }
+};
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+constexpr int N = 1e7 + 10;
+int min_fact[N];
+void prep() {
+    for (int i = 2; i < N; ++i) {
+        for (int j = i; j < N; j += i) {
+            if (not min_fact[j]) min_fact[j] = i;
+        }
+    }
+}
+
+void solve() {
+    read(int, x, y);
+    int diff = abs(x - y);
+    int res = INF;
+    while (diff != 1) {
+        res = min(res, min_fact[diff] - x % min_fact[diff]);
+        diff /= min_fact[diff];
+    }
+    if (gcd(x, y) != 1) res = 0;
+    cout << (res == INF ? -1 : 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 (t < (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1771c.cc

@@ -0,0 +1,504 @@
+#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) remove_reference<decltype(container)>::type
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+using ld = long double;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+using pid = pair<int, ld>;
+using pdi = pair<ld, int>;
+using pld = pair<ll, ld>;
+using pdl = pair<ld, ll>;
+using pdd = pair<ld, ld>;
+using tlll = tuple<ll, ll, ll>;
+using tlld = tuple<ll, ll, ld>;
+using tlli = tuple<ll, ll, int>;
+using tldl = tuple<ll, ld, ll>;
+using tldd = tuple<ll, ld, ld>;
+using tldi = tuple<ll, ld, int>;
+using tlil = tuple<ll, int, ll>;
+using tlid = tuple<ll, int, ld>;
+using tlii = tuple<ll, int, int>;
+using tdll = tuple<ld, ll, ll>;
+using tdld = tuple<ld, ll, ld>;
+using tdli = tuple<ld, ll, int>;
+using tddl = tuple<ld, ld, ll>;
+using tddd = tuple<ld, ld, ld>;
+using tddi = tuple<ld, ld, int>;
+using tdil = tuple<ld, int, ll>;
+using tdid = tuple<ld, int, ld>;
+using tdii = tuple<ld, int, int>;
+using till = tuple<int, ll, ll>;
+using tild = tuple<int, ll, ld>;
+using tili = tuple<int, ll, int>;
+using tidl = tuple<int, ld, ll>;
+using tidd = tuple<int, ld, ld>;
+using tidi = tuple<int, ld, int>;
+using tiil = tuple<int, int, ll>;
+using tiid = tuple<int, int, ld>;
+using tiii = tuple<int, int, int>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define 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 edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w), ch[v].emplace_back(u, w);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w);)
+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 (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+
+/* 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) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    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; }
+};
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+void solve() {
+    read(int, 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 (t < (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1775d.cc

@@ -0,0 +1,570 @@
+#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) remove_reference<decltype(container)>::type
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+using ld = long double;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+using pid = pair<int, ld>;
+using pdi = pair<ld, int>;
+using pld = pair<ll, ld>;
+using pdl = pair<ld, ll>;
+using pdd = pair<ld, ld>;
+using tlll = tuple<ll, ll, ll>;
+using tlld = tuple<ll, ll, ld>;
+using tlli = tuple<ll, ll, int>;
+using tldl = tuple<ll, ld, ll>;
+using tldd = tuple<ll, ld, ld>;
+using tldi = tuple<ll, ld, int>;
+using tlil = tuple<ll, int, ll>;
+using tlid = tuple<ll, int, ld>;
+using tlii = tuple<ll, int, int>;
+using tdll = tuple<ld, ll, ll>;
+using tdld = tuple<ld, ll, ld>;
+using tdli = tuple<ld, ll, int>;
+using tddl = tuple<ld, ld, ll>;
+using tddd = tuple<ld, ld, ld>;
+using tddi = tuple<ld, ld, int>;
+using tdil = tuple<ld, int, ll>;
+using tdid = tuple<ld, int, ld>;
+using tdii = tuple<ld, int, int>;
+using till = tuple<int, ll, ll>;
+using tild = tuple<int, ll, ld>;
+using tili = tuple<int, ll, int>;
+using tidl = tuple<int, ld, ll>;
+using tidd = tuple<int, ld, ld>;
+using tidi = tuple<int, ld, int>;
+using tiil = tuple<int, int, ll>;
+using tiid = tuple<int, int, ld>;
+using tiii = tuple<int, int, int>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define 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 edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w), ch[v].emplace_back(u, w);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w);)
+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 (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+
+/* 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) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    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; }
+};
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+void solve() {
+    read(int, n);
+    unordered_map<int, unordered_map<int, int, safe_hash>, safe_hash> ch;
+    readvec1(int, a, n);
+    read(int, s, t);
+    deque<tuple<int, int, int>> q;
+    unordered_set<int, safe_hash> tar;
+    for (int i = 1; i <= n; ++i) {
+        auto f = decompose_prime(a[i]);
+        int m = f.size();
+        for (int j = 0; j < m; ++j) {
+            for (int k = j + 1; k < m; ++k) {
+                if (not ch[f[j].first].count(f[k].first) or t == i) ch[f[j].first][f[k].first] = i;
+                if (not ch[f[k].first].count(f[j].first) or t == i) ch[f[k].first][f[j].first] = i;
+            }
+        }
+        if (s == i) {
+            for (auto&& [x, _] : f) {
+                q.emplace_back(x, -1, -1);
+            }
+        }
+        if (t == i) {
+            for (auto&& [x, _] : f) {
+                tar.emplace(x);
+            }
+        }
+    }
+    if (s == t) {
+        cout << 1 << '\n';
+        cout << s << '\n';
+        return;
+    }
+    // if (gcd(a[s], a[t]) != 1) {
+    //     cout << 2 << '\n';
+    //     cout << s << ' ' << t << '\n';
+    //     return;
+    // }
+    int res = INF;
+    unordered_set<int, safe_hash> vis;
+    unordered_map<int, pii, safe_hash> father;
+    while (q.size()) {
+        popfront(q, v, f, e);
+        if (vis.count(v)) continue;
+        vis.emplace(v);
+        father[v] = {f, e};
+        if (tar.count(v)) {
+            res = v;
+            break;
+        }
+        for (auto&& [u, h] : ch[v]) {
+            q.emplace_back(u, v, h);
+        }
+    }
+    if (res == INF) {
+        cout << -1 << '\n';
+    } else {
+        vector<int> path;
+        if (father[res].second != t) {
+            path.emplace_back(t);
+        }
+        while (father[res].first != -1) {
+            path.emplace_back(father[res].second);
+            res = father[res].first;
+        }
+        path.emplace_back(s);
+        reverse(path.begin(), path.end());
+        cout << path.size() << '\n';
+        putvec(path);
+    }
+}
+
+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 (t < (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1781d.cc

@@ -0,0 +1,527 @@
+#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) remove_reference<decltype(container)>::type
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+using ld = long double;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+using pid = pair<int, ld>;
+using pdi = pair<ld, int>;
+using pld = pair<ll, ld>;
+using pdl = pair<ld, ll>;
+using pdd = pair<ld, ld>;
+using tlll = tuple<ll, ll, ll>;
+using tlld = tuple<ll, ll, ld>;
+using tlli = tuple<ll, ll, int>;
+using tldl = tuple<ll, ld, ll>;
+using tldd = tuple<ll, ld, ld>;
+using tldi = tuple<ll, ld, int>;
+using tlil = tuple<ll, int, ll>;
+using tlid = tuple<ll, int, ld>;
+using tlii = tuple<ll, int, int>;
+using tdll = tuple<ld, ll, ll>;
+using tdld = tuple<ld, ll, ld>;
+using tdli = tuple<ld, ll, int>;
+using tddl = tuple<ld, ld, ll>;
+using tddd = tuple<ld, ld, ld>;
+using tddi = tuple<ld, ld, int>;
+using tdil = tuple<ld, int, ll>;
+using tdid = tuple<ld, int, ld>;
+using tdii = tuple<ld, int, int>;
+using till = tuple<int, ll, ll>;
+using tild = tuple<int, ll, ld>;
+using tili = tuple<int, ll, int>;
+using tidl = tuple<int, ld, ll>;
+using tidd = tuple<int, ld, ld>;
+using tidi = tuple<int, ld, int>;
+using tiil = tuple<int, int, ll>;
+using tiid = tuple<int, int, ld>;
+using tiii = tuple<int, int, int>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define 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 edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w), ch[v].emplace_back(u, w);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w);)
+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 (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+
+/* 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) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
+    }
+    return res;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    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; }
+};
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+void solve() {
+    read(int, n);
+    readvec(int, a, n);
+    vector<unordered_map<ll, int, safe_hash>> choices(n);
+    for (int i = 0; i < n; ++i) {
+        for (int j = i + 1; j < n; ++j) {
+            int diff = a[j] - a[i];
+            int sq = sqrt(diff);
+            for (int j = 1; j <= sq; ++j) {
+                if (diff % j == 0 and ((diff / j) & 1) == (j & 1)) {
+                    ll x = j, y = diff / j;
+                    ll q = x + y >> 1, p = y - q;
+                    if (p * p >= a[i]) {
+                        choices[i][p] += 1;
+                    }
+                }
+            }
+        }
+    }
+    int res = 0;
+    for (auto&& d : choices) {
+        for (auto&& [_, v] : d) {
+            res = max(res, v);
+        }
+    }
+    cout << res + 1 << '\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 (t < (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/ctext_out.cc

@@ -381,17 +381,42 @@ template <ll mdl> struct MLL {
     void operator/=(const MLL& rhs) { val = (*this / rhs).val; }
     void operator%=(const MLL& rhs) { val = (*this % rhs).val; }
 };
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
 
 template <ll mdl>
 ostream& operator<<(ostream& out, const MLL<mdl>& num) {
     return out << num.val;
 }
 
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
 template <ll mdl>
 istream& operator>>(istream& in, MLL<mdl>& num) {
     return in >> num.val;
 }
 
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
 // miscancellous
 template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
     std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
@@ -444,54 +469,35 @@ void dump() {}
 
 void dump_ignore() {}
 
-void prep() {}
+void prep() {
+}
 
 void solve() {
-    read(int, n, m);
-    vector<vector<int>> a(n, vector<int>(m + 1));
-    for (int i = 0; i < n; ++i) {
-        for (int j = 1; j <= m; ++j) {
-            cin >> a[i][j];
+    read(int, n);
+    readvec(int, a, n);
+    vector<int> pos;
+    for (int i = 1; i < n; i += 2) {
+        if (a[i] == 1) {
+            i += 1;
         }
+        if (i == n) break;
+        pos.emplace_back(i);
     }
+    int target = n;
     vector<int> res(n);
-    auto rev = [&] (const vector<int>& x) {
-        vector<int> res(m + 1);
-        for (int i = 0; i <= m; ++i) {
-            int curr = x[i];
-            res[curr] = i;
+    sort_by_key(pos.begin(), pos.end(), [&] (int i) { return a[i]; });
+    for (auto&& i : pos) {
+        res[i] = target--;
     }
-        return res;
-    };
-    vector<array<int, 11>> trie(1);
-    auto insert = [&] (const vector<int>& x) -> void {
-        int curr = 0;
-        for (int i = 1; i <= m; ++i) {
-            // debug(i), debug(x[i]);
-            // debug(curr);
-            if (not trie[curr][x[i]]) {
-                trie[curr][x[i]] = trie.size();
-                trie.push_back({});
-            }
-            curr = trie[curr][x[i]];
-        }
-    };
-    auto query = [&] (const vector<int>& x) -> int {
-        int curr = 0;
-        for (int i = 1; i <= m; ++i) {
-            if (not trie[curr][x[i]]) {
-                return i - 1;
-            }
-            curr = trie[curr][x[i]];
-        }
-        return m;
-    };
+    vector<int> rem;
     for (int i = 0; i < n; ++i) {
-        insert(rev(a[i]));
+        if (not res[i]) rem.emplace_back(i);
     }
-    for (int i = 0; i < n; ++i) {
-        cout << query(a[i]) << " \n"[i + 1 == n];
+    sort_by_key(rem.begin(), rem.end(), [&] (int i) { return a[i]; });
+    for (auto&& i : rem) {
+        res[i] = target--;
     }
+    putvec(res);
 }
 
 int main() {
@@ -506,7 +512,7 @@ int main() {
     read(int, t);
     for (int i = 0; i < t; ++i) {
 #ifdef DUMP_TEST_CASE
-        if (t < (TOT_TEST_CASE)) {
+        if (t != (TOT_TEST_CASE)) {
             solve();
         } else if (i + 1 == (DUMP_TEST_CASE)) {
             dump();

src/bin/d.cc

@@ -15,7 +15,7 @@ using namespace std;
 #define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
 constexpr void __() {}
 #define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
-#define __as_typeof(container) decltype(container)::value_type
+#define __as_typeof(container) remove_reference<decltype(container)>::type
 
 /* type aliases */
 #if LONG_LONG_MAX != INT64_MAX
@@ -67,6 +67,8 @@ using tiid = tuple<int, int, ld>;
 using tiii = tuple<int, int, int>;
 template <typename T> using max_heap = priority_queue<T>;
 template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
 
 /* constants */
 constexpr int INF = 0x3f3f3f3f;
@@ -201,10 +203,17 @@ template <typename T, typename Iterator> pair<size_t, unordered_map<T, size_t, s
 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 readvec(type, a, n) __AS_PROCEDURE(vector<type> a(n); for (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
 #define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
 #define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
 template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
     out << "{" << p.first << ", " << p.second << "}";
     return out;
@@ -372,17 +381,42 @@ template <ll mdl> struct MLL {
     void operator/=(const MLL& rhs) { val = (*this / rhs).val; }
     void operator%=(const MLL& rhs) { val = (*this % rhs).val; }
 };
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
 
 template <ll mdl>
 ostream& operator<<(ostream& out, const MLL<mdl>& num) {
     return out << num.val;
 }
 
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
 template <ll mdl>
 istream& operator>>(istream& in, MLL<mdl>& num) {
     return in >> num.val;
 }
 
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
 // miscancellous
 template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
     std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
@@ -390,45 +424,98 @@ template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, Ran
 template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
     std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
 }
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
 /////////////////////////////////////////////////////////
 
-#define SINGLE_TEST_CASE
+// #define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
 
 void dump() {}
 
 void dump_ignore() {}
 
-void prep() {}
+void prep() {
+}
 
 void solve() {
-    using mll = MLL<PRIME>;
-    vector<mll> pw(20);
-    pw[0] = 1;
-    for (int i = 1; i < 20; ++i) {
-        pw[i] = pw[i - 1] * 10;
-    }
-    read(int, n);
-    readvec(int, a, n);
-    array<int, 11> cnt {};
-    mll res = 0;
-    for (int i = n - 1; ~i; --i) {
-        for (int j = 1; j <= 10; ++j) {
-            res += a[i] * pw[j] * cnt[j];
-        }
-        res += mll(1) * a[i] * i;
-        int curr = 0;
-        if (a[i] == 0) {
-            curr = 1;
-        } else {
-            while (a[i]) {
-                curr += 1;
-                a[i] /= 10;
+    read(int, n, k);
+    int mx = -1;
+    for (int i = 1; i <= n; ++i) {
+        cout << "? 1 " << ll(1) * i * n << endl;
+        read(int, r);
+        if (r == n) {
+            mx = i;
+            break;
         }
     }
-        cnt[curr] += 1;
+    if (mx == -1) exit(825);
+    for (int i = n / k; i >= 1; --i) {
+        ll m = ll(1) * mx * i;
+        int prev = 1;
+        int f = 1;
+        for (int j = 1; j <= k; ++j) {
+            if (prev == n + 1) {
+                f = 0;
+                break;
+            }
+            cout << "? " << prev << ' ' << m << endl;
+            read(int, r);
+            if (r == n + 1) {
+                f = 0;
+                break;
+            }
+            prev = r + 1;
+        }
+        if (f and prev == n + 1) {
+            // ok
+            cout << "! " << m << endl;
+            read(int, stat);
+            if (stat == -1) {
+                exit(0);
+            }
+            return;
+        }
+    }
+    cout << "! -1" << endl;
+    read(int, stat);
+    if (stat == -1) {
+        exit(0);
     }
-    cout << res << '\n';
 }
 
 int main() {
@@ -443,7 +530,7 @@ int main() {
     read(int, t);
     for (int i = 0; i < t; ++i) {
 #ifdef DUMP_TEST_CASE
-        if (t < (DUMP_TEST_CASE)) {
+        if (t != (TOT_TEST_CASE)) {
             solve();
         } else if (i + 1 == (DUMP_TEST_CASE)) {
             dump();

src/bin/std.in

@@ -1,18 +1,9 @@
-3
-3 4
-2 4 1 3
-1 2 4 3
-2 1 3 4
-2 2
-1 2
-2 1
-8 10
-3 4 9 6 10 2 7 8 1 5
-3 9 1 8 5 7 4 10 2 6
-3 10 1 7 5 9 6 4 2 8
-1 2 3 4 8 6 10 7 9 5
-1 2 3 4 10 6 8 5 7 9
-9 6 1 2 10 4 7 8 3 5
-7 9 3 2 5 6 4 8 1 10
-9 4 3 7 5 6 1 10 8 2
-
+4
+4
+1 2 3 4
+4
+4 3 1 2
+6
+6 5 1 4 2 3
+8
+1 2 4 5 7 6 8 3

src/bin/template.cc

@@ -381,17 +381,42 @@ template <ll mdl> struct MLL {
     void operator/=(const MLL& rhs) { val = (*this / rhs).val; }
     void operator%=(const MLL& rhs) { val = (*this % rhs).val; }
 };
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
 
 template <ll mdl>
 ostream& operator<<(ostream& out, const MLL<mdl>& num) {
     return out << num.val;
 }
 
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
 template <ll mdl>
 istream& operator>>(istream& in, MLL<mdl>& num) {
     return in >> num.val;
 }
 
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
 // miscancellous
 template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
     std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
@@ -436,7 +461,7 @@ public:
 };
 /////////////////////////////////////////////////////////
 
-#define SINGLE_TEST_CASE
+// #define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -444,7 +469,8 @@ void dump() {}
 
 void dump_ignore() {}
 
-void prep() {}
+void prep() {
+}
 
 void solve() {
 }
@@ -461,7 +487,7 @@ int main() {
     read(int, t);
     for (int i = 0; i < t; ++i) {
 #ifdef DUMP_TEST_CASE
-        if (t < (TOT_TEST_CASE)) {
+        if (t != (TOT_TEST_CASE)) {
             solve();
         } else if (i + 1 == (DUMP_TEST_CASE)) {
             dump();

src/bin/test.cc

@@ -1,16 +1,500 @@
-#include <bits/stdc++.h>
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * Useful Macros
+ *   by subcrip
+ * (requires C++17)
+ */
+
+#include<bits/stdc++.h>
 using namespace std;
 
-int main() {
-    int m = 4;
-    auto rev = [&] (const vector<int>& x) {
-        vector<int> res(m + 1);
-        for (int i = 0; i < m; ++i) {
-            int curr = x[i];
-            res[curr] = i + 1;
+/* macro helpers */
+#define __NARGS(...) std::tuple_size<decltype(std::make_tuple(__VA_ARGS__))>::value
+#define __DECOMPOSE_S(a, x) auto x = a;
+#define __DECOMPOSE_N(a, ...) auto [__VA_ARGS__] = a;
+constexpr void __() {}
+#define __AS_PROCEDURE(...) __(); __VA_ARGS__; __()
+#define __as_typeof(container) remove_reference<decltype(container)>::type
+
+/* type aliases */
+#if LONG_LONG_MAX != INT64_MAX
+using ll = int64_t;
+using ull = uint64_t;
+#else
+using ll = long long;
+using ull = unsigned long long;
+using ld = long double;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;
+using pil = pair<int, ll>;
+using pli = pair<ll, int>;
+using pll = pair<ll, ll>;
+using pid = pair<int, ld>;
+using pdi = pair<ld, int>;
+using pld = pair<ll, ld>;
+using pdl = pair<ld, ll>;
+using pdd = pair<ld, ld>;
+using tlll = tuple<ll, ll, ll>;
+using tlld = tuple<ll, ll, ld>;
+using tlli = tuple<ll, ll, int>;
+using tldl = tuple<ll, ld, ll>;
+using tldd = tuple<ll, ld, ld>;
+using tldi = tuple<ll, ld, int>;
+using tlil = tuple<ll, int, ll>;
+using tlid = tuple<ll, int, ld>;
+using tlii = tuple<ll, int, int>;
+using tdll = tuple<ld, ll, ll>;
+using tdld = tuple<ld, ll, ld>;
+using tdli = tuple<ld, ll, int>;
+using tddl = tuple<ld, ld, ll>;
+using tddd = tuple<ld, ld, ld>;
+using tddi = tuple<ld, ld, int>;
+using tdil = tuple<ld, int, ll>;
+using tdid = tuple<ld, int, ld>;
+using tdii = tuple<ld, int, int>;
+using till = tuple<int, ll, ll>;
+using tild = tuple<int, ll, ld>;
+using tili = tuple<int, ll, int>;
+using tidl = tuple<int, ld, ll>;
+using tidd = tuple<int, ld, ld>;
+using tidi = tuple<int, ld, int>;
+using tiil = tuple<int, int, ll>;
+using tiid = tuple<int, int, ld>;
+using tiii = tuple<int, int, int>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define 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 edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w), ch[v].emplace_back(u, w);)
+#define Edge(ch, u, v) __AS_PROCEDURE(ch[u].push_back(v);)
+#define Edgew(ch, u, v, w) __AS_PROCEDURE(ch[u].emplace_back(v, w);)
+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 (auto& x : a) cin >> x;)
+#define readvec1(type, a, n) __AS_PROCEDURE(vector<type> a((n) + 1); copy_n(ii<type>(cin), (n), a.begin() + 1);)
+#define putvec(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec1(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, " ")); cout << endl;)
+#define putvec_eol(a) __AS_PROCEDURE(copy(a.begin(), a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
+#define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
+#define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
+    return in >> p.first >> p.second;
+}
+template<typename T, typename U> ostream& operator<<(ostream& out, const pair<T, U>& p) {
+    out << "{" << p.first << ", " << p.second << "}";
+    return out;
+}
+template<typename Char, typename Traits, typename Tuple, std::size_t... Index>
+void print_tuple_impl(std::basic_ostream<Char, Traits>& os, const Tuple& t, std::index_sequence<Index...>) {
+    using swallow = int[]; // guaranties left to right order
+    (void)swallow { 0, (void(os << (Index == 0 ? "" : ", ") << std::get<Index>(t)), 0)... };
+}
+template<typename Char, typename Traits, typename... Args>
+decltype(auto) operator<<(std::basic_ostream<Char, Traits>& os, const std::tuple<Args...>& t) {
+    os << "{";
+    print_tuple_impl(os, t, std::index_sequence_for<Args...>{});
+    return os << "}";
+}
+template<typename T> ostream& operator<<(ostream& out, const vector<T>& vec) {
+    for (auto&& i : vec) out << i << ' ';
+    return out;
+}
+std::ostream& operator<<(std::ostream& dest, const int128& value) {
+    // https://stackoverflow.com/a/25115163/23881100
+    std::ostream::sentry s( dest );
+    if ( s ) {
+        uint128 tmp = value < 0 ? -value : value;
+        char buffer[ 128 ];
+        char* d = std::end( buffer );
+        do {
+            -- d;
+            *d = "0123456789"[ tmp % 10 ];
+            tmp /= 10;
+        } while ( tmp != 0 );
+        if ( value < 0 ) {
+            -- d;
+            *d = '-';
+        }
+        int len = std::end( buffer ) - d;
+        if ( dest.rdbuf()->sputn( d, len ) != len ) {
+            dest.setstate( std::ios_base::badbit );
+        }
+    }
+    return dest;
+}
+
+/* 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) {
+    // return (factor, count, factor ** count)
+    vector<tuple<int, int, ll>> res;
+    for (int i = 2; i * i <= x; i++) {
+        if (x % i == 0) {
+            int cnt = 0;
+            ll pw = 1;
+            while (x % i == 0) ++cnt, x /= i, pw *= i;
+            res.emplace_back(i, cnt, pw);
+        }
+    }
+    if (x != 1) {
+        res.emplace_back(x, 1, x);
     }
     return res;
-    };
-    auto res = rev({2, 1, 3, 4});
-    for (auto&& x : res)cout << x << endl;
+}
+
+vector<pii> decompose_prime(int N) {
+    // return (factor, count)
+    vector<pii> result;
+    for (int i = 2; i * i <= N; i++) {
+        if (N % i == 0) {
+            int cnt = 0;
+            while (N % i == 0) N /= i, ++cnt;
+            result.emplace_back(i, cnt);
+        }
+    }
+    if (N != 1) {
+        result.emplace_back(N, 1);
+    }
+    return result;
+}
+
+/* string algorithms */
+vector<int> calc_next(string t) {  // pi function of t
+  int n = (int)t.length();
+  vector<int> pi(n);
+  for (int i = 1; i < n; i++) {
+    int j = pi[i - 1];
+    while (j > 0 && t[i] != t[j]) j = pi[j - 1];
+    if (t[i] == t[j]) j++;
+    pi[i] = j;
+  }
+  return pi;
+}
+vector<int> calc_z(string t) {  // z function of t
+    int m = t.length();
+    vector<int> z;
+    z.push_back(m);
+    pair<int, int> prev = {1, -1};
+    for (int i = 1; i < m; ++i) {
+        if (z[i - prev.first] + i <= prev.second) {
+            z.push_back(z[i - prev.first]);
+        } else {
+            int j = max(i, prev.second + 1);
+            while (j < m && t[j] == t[j - i]) ++j;
+            z.push_back(j - i);
+            prev = {i, j - 1};
+        }
+    }
+    return z;
+}
+vector<int> kmp(string s, string t) {  // find all t in s
+  string cur = t + '#' + s;
+  int sz1 = s.size(), sz2 = t.size();
+  vector<int> v;
+  vector<int> lps = calc_next(cur);
+  for (int i = sz2 + 1; i <= sz1 + sz2; i++) {
+    if (lps[i] == sz2) v.push_back(i - 2 * sz2);
+  }
+  return v;
+}
+int period(string s) {  // find the length of shortest recurring period
+    int n = s.length();
+    auto z = calc_z(s);
+    for (int i = 1; i <= n / 2; ++i) {
+        if (n % i == 0 && z[i] == n - i) {
+            return i;
+        }
+    }
+    return n;
+}
+
+/* modular arithmetic */
+template <ll mdl> struct MLL {
+    ll val;
+    MLL(ll v = 0) : val(mod(v, mdl)) {}
+    MLL(const MLL<mdl>& other) : val(other.val) {}
+    friend MLL operator+(const MLL& lhs, const MLL& rhs) { return mod(lhs.val + rhs.val, mdl); }
+    friend MLL operator-(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - rhs.val, mdl); }
+    friend MLL operator*(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * rhs.val, mdl); }
+    friend MLL operator/(const MLL& lhs, const MLL& rhs) { return mod(lhs.val * mod(inverse(rhs.val, mdl), mdl), mdl); }
+    friend MLL operator%(const MLL& lhs, const MLL& rhs) { return mod(lhs.val - (lhs / rhs).val, mdl); }
+    friend bool operator==(const MLL& lhs, const MLL& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLL& lhs, const MLL& rhs) { return lhs.val != rhs.val; }
+    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; }
+};
+struct MLLd {
+    ll val, mdl;
+    MLLd(ll mdl, ll v = 0) : mdl(mdl), val(mod(v, mdl)) {}
+    MLLd(const MLLd& other) : val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val + rhs.val, lhs.mdl); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - rhs.val, lhs.mdl); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * rhs.val, lhs.mdl); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return mod(lhs.val - (lhs / rhs).val, lhs.mdl); }
+    friend bool operator==(const MLLd& lhs, const MLLd& rhs) { return lhs.val == rhs.val; }
+    friend bool operator!=(const MLLd& lhs, const MLLd& rhs) { return lhs.val != rhs.val; }
+    void operator+=(const MLLd& rhs) { val = (*this + rhs).val; }
+    void operator-=(const MLLd& rhs) { val = (*this - rhs).val; }
+    void operator*=(const MLLd& rhs) { val = (*this * rhs).val; }
+    void operator/=(const MLLd& rhs) { val = (*this / rhs).val; }
+    void operator%=(const MLLd& rhs) { val = (*this % rhs).val; }
+};
+
+template <ll mdl>
+ostream& operator<<(ostream& out, const MLL<mdl>& num) {
+    return out << num.val;
+}
+
+ostream& operator<<(ostream& out, const MLLd& num) {
+    return out << num.val;
+}
+
+template <ll mdl>
+istream& operator>>(istream& in, MLL<mdl>& num) {
+    return in >> num.val;
+}
+
+istream& operator>>(istream& in, MLLd& num) {
+    return in >> num.val;
+}
+
+// miscancellous
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; a_it != a_last and b_it != b_last; ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+constexpr int MAXN = 1e6 + 10;
+using mll = MLL<MDL>;
+mll fact[MAXN], pw2[MAXN], pw3[MAXN];
+void prep() {
+    fact[0] = 1, pw2[0] = 1, pw3[0] = 1;
+    for (int i = 1; i < MAXN; ++i) {
+        fact[i] = fact[i - 1] * i;
+        pw3[i] = pw3[i - 1] * 3;
+        pw2[i] = pw2[i - 1] * 2;
+    }
+}
+
+mll comb(int n, int k) {
+    if (n < 0 or k < 0 or n < k) return 0;
+    return fact[n] / fact[k] / fact[n - k];
+}
+
+int main() {
+    prep();
+    int t = 10;
+    array<mll, 10> s;
+    for (int i = 0; i < 10; ++i) {
+        s[i] = 0;
+        for (int j = 0; j <= i; ++j) {
+            s[i] += comb(i, j) * pw3[t - j] * comb(t, j);
+        }
+        if (i >= 2) assert(s[i] == 1 + 10 * s[i - 1] / 9 - (i - 1) * comb(t, i - 2) / pw3[i]);
+    }
 }

src/bin/test.py

@@ -1,21 +1,3 @@
-from random import randint
-from os import system
-
-while 1:
-    n = 5
-    ipt = str(n) + '\n'
-    for i in range(2, n + 1):
-        ipt += f'{randint(1, i - 1)} {i}\n'
-    with open('std.in', 'w') as f:
-        f.write(ipt)
-    system('./bad < std.in > std.out')
-    with open('std.out') as f:
-        bad = f.read().strip()
-    system('./good < std.in > std.out')
-    with open('std.out') as f:
-        good = f.read().strip()
-    if bad != good:
-        print('input:', ipt)
-        print('bad:', bad)
-        print('good:', good)
-        break
+print(300000)
+print('277200 ' * 300000)
+print('1 200')