backup

src/bin/.nvim.lua

@@ -1,7 +1,7 @@
 vim.api.nvim_create_autocmd("FileType", {
     pattern = "cpp",
     callback = function()
-        vim.api.nvim_buf_set_keymap(0, 'n', "<leader>b", '<Cmd>10sp<CR><Cmd>te ctext -i $(gcc -xc++ /dev/null -E -Wp,-v 2>&1 | sed -n "s,^ ,,p" | tr "\\n" ",") < % > ctext_out.cc && clang++ -std=c++17 -DONLINE_JUDGE -Wl,-z,stack-size=268435456 -Wall -Ofast -g -fsanitize=address -fsanitize=undefined ctext_out.cc && ./a.out < std.in<CR>i', {
+        vim.api.nvim_buf_set_keymap(0, 'n', "<leader>b", '<Cmd>10sp<CR><Cmd>te ctext -i $(gcc -xc++ /dev/null -E -Wp,-v 2>&1 | sed -n "s,^ ,,p" | tr "\\n" ",") < % > ctext_out.cc && clang++ -std=c++17 -DONLINE_JUDGE -Wl,-z,stack-size=268435456 -Wall -Ofast -g -fsanitize=address -fsanitize=undefined ctext_out.cc && time ./a.out < std.in<CR>i', {
             silent = true,
             noremap = true
         })

src/bin/a.cc

@@ -23,47 +23,22 @@ 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 pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
-using pil = pair<int, ll>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
-using pli = pair<ll, int>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
-using pll = pair<ll, ll>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
-using pid = pair<int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
-using pdi = pair<ld, int>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
-using pld = pair<ll, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
-using pdl = pair<ld, ll>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
-using pdd = pair<ld, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
-using tlll = tuple<ll, ll, ll>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
-using tlld = tuple<ll, ll, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
-using tlli = tuple<ll, ll, int>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
-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>;
@@ -211,6 +186,7 @@ template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __
 #define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
 #define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
 #define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
 template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
     return in >> p.first >> p.second;
 }
@@ -429,6 +405,23 @@ istream& operator>>(istream& in, MLLd& num) {
 }
 
 // miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
 #define functor(func) [&](auto&&... val) \
 noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
 {return func(std::forward<decltype(val)>(val)...);}
@@ -487,7 +480,7 @@ array<T, N> __initarray(const T& init) {
 }
 /////////////////////////////////////////////////////////
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -500,8 +493,21 @@ void prep() {
 
 void solve() {
     read(int, n);
-    readvec(int, a, n);
+    read(string, a);
-    cout << a[n - 1] + *max_element(a.begin(), a.begin() + n - 1) << '\n';
+
+    using mll = MLL<MDL>;
+    mll res = 1;
+    int cnt = 1;
+    for (int i = 1; i < n; ++i) {
+        if (a[i] != a[i - 1]) {
+            cnt += 1;
+        } else {
+            res *= (cnt + 1) / 2;
+            cnt = 1;
+        }
+    }
+    res *= (cnt + 1) / 2;
+    cout << res << '\n';
 }
 
 int main() {

src/bin/abc-355f.cc

@@ -0,0 +1,590 @@
+#pragma GCC diagnostic ignored "-Wunused-const-variable"
+#pragma GCC diagnostic ignored "-Wreorder-ctor"
+#pragma GCC diagnostic ignored "-Wunknown-pragmas"
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+class quick_union {
+private:
+    vector<size_t> c, sz;
+public:
+    quick_union(size_t n) : c(n), sz(n) {
+        iota(c.begin(), c.end(), 0);
+        sz.assign(n, 1);
+    }
+
+    size_t query(size_t i) {
+        if (c[i] != i) c[i] = query(c[i]);
+        return c[i];
+    }
+
+    void merge(size_t i, size_t j) {
+        if (connected(i, j)) return;
+        sz[query(j)] += sz[query(i)];
+        c[query(i)] = query(j);
+    }
+    bool connected(size_t i, size_t j) {
+        return query(i) == query(j);
+    }
+    size_t query_size(size_t i) {
+        return sz[query(i)];
+    }
+};
+
+void solve() {
+    read(int, n, q);
+    vector<tiii> edges;
+    int base = 0;
+    for (int i = 0; i < n - 1; ++i) {
+        read(int, u, v, w);
+        edges.emplace_back(w, u, v);
+        base += w;
+    }
+    sort(edges.begin(), edges.end());
+
+    vector<tiii> qe;
+    for (int i = 0; i < q; ++i) {
+        read(int, u, v, w);
+        qe.emplace_back(w, u, v);
+    }
+
+    vector<int> res(q, INF);
+    for (int i = 1; i <= 10; ++i) {
+        quick_union qu(n + 1);
+        for (auto&& [w, u, v] : edges) {
+            if (w > i) break;
+            qu.merge(u, v);
+        }
+        for (int j = 0; j < q; ++j) {
+            auto&& [w, u, v] = qe[j];
+            if (qu.connected(u, v) and res[j] == INF) {
+                res[j] = min(0, w - i);
+            }
+            if (w <= i) {
+                qu.merge(u, v);
+            }
+        }
+    }
+
+    assert(count(res.begin(), res.end(), INF) == 0);
+    int curr = base;
+    for (int i = 0; i < q; ++i) {
+        cout << (curr += res[i]) << '\n';
+    }
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/abc-357f.cc

@@ -0,0 +1,663 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+using mll = MLL<PRIME>;
+
+template<typename Addable_Info_t, typename Tag_t, typename Sequence = std::vector<Addable_Info_t>> class segtree {
+private:
+    using size_type = uint64_t;
+    using info_type = Addable_Info_t;
+    using tag_type = Tag_t;
+    size_type _max;
+    vector<info_type> d;
+    vector<tag_type> b;
+    void pull(size_type p) {
+        d[p] = d[p * 2] + d[p * 2 + 1];
+    }
+    void push(size_type p, size_type left_len, size_type right_len) {
+        d[p * 2].apply(b[p], left_len), d[p * 2 + 1].apply(b[p], right_len);
+        b[p * 2].apply(b[p]), b[p * 2 + 1].apply(b[p]);
+        b[p] = tag_type();
+    }
+    void set(size_type s, size_type t, size_type p, size_type x, const info_type& c) {
+        if (s == t) {
+            d[p] = c;
+            return;
+        }
+        size_type m = s + (t - s >> 1);
+        if (s != t) push(p, m - s + 1, t - m);
+        if (x <= m) set(s, m, p * 2, x, c);
+        else set(m + 1, t, p * 2 + 1, x, c);
+        pull(p);
+    }
+
+    void range_apply(size_type s, size_type t, size_type p, size_type l, size_type r, const tag_type& c) {
+        if (l <= s && t <= r) {
+            d[p].apply(c, t - s + 1);
+            b[p].apply(c);
+            return;
+        }
+        size_type m = s + (t - s >> 1);
+        push(p, m - s + 1, t - m);
+        if (l <= m) range_apply(s, m, p * 2, l, r, c);
+        if (r > m)  range_apply(m + 1, t, p * 2 + 1, l, r, c);
+        pull(p);
+    }
+    info_type range_query(size_type s, size_type t, size_type p, size_type l, size_type r) {
+        if (l <= s && t <= r) {
+            return d[p];
+        }
+        size_type m = s + (t - s >> 1);
+        info_type res = {};
+        push(p, m - s + 1, t - m);
+        if (l <= m) res = res + range_query(s, m, p * 2, l, r);
+        if (r > m)  res = res + range_query(m + 1, t, p * 2 + 1, l, r);
+        return res;
+    }
+    void build(const Sequence& a, size_type s, size_type t, size_type p) {
+        if (s == t) {
+            d[p] = a[s];
+            return;
+        }
+        int m = s + (t - s >> 1);
+        build(a, s, m, p * 2);
+        build(a, m + 1, t, p * 2 + 1);
+        pull(p);
+    }
+public:
+    segtree(size_type __max) : d(4 * __max), b(4 * __max), _max(__max - 1) {}
+    segtree(const Sequence& a) : segtree(a.size()) {
+        build(a, {}, _max, 1);
+    }
+    void set(size_type i, const info_type& c) {
+        set({}, _max, 1, i, c);
+    }
+
+    void range_apply(size_type l, size_type r, const tag_type& c) {
+        range_apply({}, _max, 1, l, r, c);
+    }
+    void apply(size_type i, const tag_type& c) {
+        range_apply(i, i, c);
+    }
+    info_type range_query(size_type l, size_type r) {
+        return range_query({}, _max, 1, l, r);
+    }
+    info_type query(size_type i) {
+        return range_query(i, i);
+    }
+    Sequence serialize() {
+        Sequence res = {};
+        for (size_type i = 0; i <= _max; ++i) {
+            res.push_back(query(i));
+        }
+        return res;
+    }
+    const vector<info_type>& get_d() {
+        return d;
+    }
+};
+struct Tag {
+    mll a = 0, b = 0;
+    void apply(const Tag& rhs) {
+        a += rhs.a;
+        b += rhs.b;
+    }
+};
+struct Info {
+    mll a_sum = 0, b_sum = 0, val = 0;
+    void apply(const Tag& rhs, size_t len) {
+        val += rhs.b * a_sum;
+        b_sum += rhs.b * len;
+        val += rhs.a * b_sum;
+        a_sum += rhs.a * len;
+    }
+};
+Info operator+(const Info &a, const Info &b) {
+    return {
+        .a_sum = a.a_sum + b.a_sum,
+        .b_sum = a.b_sum + b.b_sum,
+        .val = a.val + b.val,
+    };
+}
+
+void solve() {
+    read(int, n, q);
+    readvec(mll, a, n);
+    readvec(mll, b, n);
+
+
+    segtree<Info, Tag> tr(n);
+    for (int i = 0; i < n; ++i) {
+        tr.set(i, { a[i], b[i], a[i] * b[i] });
+    }
+
+    while (q--) {
+        read(int, op, l, r);
+        --l, --r;
+        if (op == 1) {
+            read(int, x);
+            tr.range_apply(l, r, { x, 0 });
+        } else if (op == 2) {
+            read(int, x);
+            tr.range_apply(l, r, { 0, x });
+        } else {
+            cout << tr.range_query(l, r).val << '\n';
+        }
+    }
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/abc-359f.cc

@@ -0,0 +1,525 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#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);
+    sort(a.begin(), a.end());
+
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/arc-180a.cc

@@ -0,0 +1,537 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#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);
+    read(string, a);
+
+    using mll = MLL<MDL>;
+    mll res = 1;
+    int cnt = 1;
+    for (int i = 1; i < n; ++i) {
+        if (a[i] != a[i - 1]) {
+            cnt += 1;
+        } else {
+            res *= (cnt + 1) / 2;
+            cnt = 1;
+        }
+    }
+    res *= (cnt + 1) / 2;
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/arc-180b.cc

@@ -0,0 +1,549 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#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);
+    readvec(int, a, n);
+    vector<pii> res;
+    while (1) {
+        int pos = -1;
+        vector<int> ss(n + 1, INF);
+        for (int i = n - 1; ~i; --i) {
+            ss[i] = min(ss[i + 1], a[i]);
+        }
+        for (int i = n - k - 1; i >= 0; --i) {
+            if (ss[i + k] < a[i] and (pos == -1 or a[i] < a[pos])) {
+                pos = i;
+            }
+        }
+        if (pos == -1) break;
+        int j = -1;
+        for (int i = pos + k; i < n; ++i) {
+            if (a[i] < a[pos] and (j == -1 or a[i] > a[j])) {
+                j = i;
+            }
+        }
+        swap(a[pos], a[j]);
+        res.emplace_back(pos + 1, j + 1);
+    }
+    cout << res.size() << '\n';
+    for (auto&& [x, y] : res) {
+        cout << x << ' ' << y << '\n';
+    }
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/arc-180c.cc

@@ -0,0 +1,552 @@
+#include <numeric>
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#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);
+    constexpr int Z = 1000;
+    using mll = MLL<MDL>;
+    vector dp(n + 1, array<mll, 2 * Z + 1>());
+
+    mll res = 1;
+    dp[0][Z] = 1;
+    unordered_set<int> oc;
+    for (int i = 1; i <= n; ++i) {
+        for (int j = 0; j <= 2 * Z; ++j) {
+            if (j - a[i - 1] == Z) {
+                res -= dp[i - 1][j - a[i - 1]];
+                if (oc.count(a[i - 1])) {
+                    ;;
+                } else {
+                    oc.emplace(a[i - 1]);
+                    dp[i][j] += dp[i - 1][j - a[i - 1]];
+                }
+            } else if (j - a[i - 1] >= 0 and j - a[i - 1] <= 2 * Z) {
+                dp[i][j] += dp[i - 1][j - a[i - 1]];
+            }
+            dp[i][j] += dp[i - 1][j];
+        }
+    }
+
+    debug(accumulate(dp[n].begin(), dp[n].end(), mll(0)));
+    res += accumulate(dp[n].begin(), dp[n].end(), mll(0));
+
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/b.cc

@@ -23,47 +23,22 @@ 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 pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
-using pil = pair<int, ll>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
-using pli = pair<ll, int>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
-using pll = pair<ll, ll>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
-using pid = pair<int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
-using pdi = pair<ld, int>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
-using pld = pair<ll, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
-using pdl = pair<ld, ll>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
-using pdd = pair<ld, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
-using tlll = tuple<ll, ll, ll>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
-using tlld = tuple<ll, ll, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
-using tlli = tuple<ll, ll, int>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
-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>;
@@ -211,6 +186,7 @@ template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __
 #define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
 #define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
 #define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
 template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
     return in >> p.first >> p.second;
 }
@@ -429,6 +405,23 @@ istream& operator>>(istream& in, MLLd& num) {
 }
 
 // miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
 #define functor(func) [&](auto&&... val) \
 noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
 {return func(std::forward<decltype(val)>(val)...);}
@@ -487,7 +480,7 @@ array<T, N> __initarray(const T& init) {
 }
 /////////////////////////////////////////////////////////
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -499,12 +492,34 @@ void prep() {
 }
 
 void solve() {
-    read(ll, n, a, b);
+    read(int, n, k);
-    auto calc = [&] (ll k) {
+    readvec(int, a, n);
-        k = max(ll(0), min(k, min(n, b)));
+    vector<pii> res;
-        return k * (b + 1) - (k + 1) * k / 2 + a * (n - k);
+    while (1) {
-    };
+        int pos = -1;
-    cout << max(calc(b - a - 1), calc(b - a)) << '\n';
+        vector<int> ss(n + 1, INF);
+        for (int i = n - 1; ~i; --i) {
+            ss[i] = min(ss[i + 1], a[i]);
+        }
+        for (int i = n - k - 1; i >= 0; --i) {
+            if (ss[i + k] < a[i] and (pos == -1 or a[i] < a[pos])) {
+                pos = i;
+            }
+        }
+        if (pos == -1) break;
+        int j = -1;
+        for (int i = pos + k; i < n; ++i) {
+            if (a[i] < a[pos] and (j == -1 or a[i] > a[j])) {
+                j = i;
+            }
+        }
+        swap(a[pos], a[j]);
+        res.emplace_back(pos + 1, j + 1);
+    }
+    cout << res.size() << '\n';
+    for (auto&& [x, y] : res) {
+        cout << x << ' ' << y << '\n';
+    }
 }
 
 int main() {

src/bin/c.cc

@@ -1,3 +1,4 @@
+#include <numeric>
 #pragma GCC optimize("Ofast")
 /////////////////////////////////////////////////////////
 /**
@@ -23,47 +24,22 @@ 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 pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
-using pil = pair<int, ll>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
-using pli = pair<ll, int>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
-using pll = pair<ll, ll>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
-using pid = pair<int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
-using pdi = pair<ld, int>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
-using pld = pair<ll, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
-using pdl = pair<ld, ll>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
-using pdd = pair<ld, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
-using tlll = tuple<ll, ll, ll>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
-using tlld = tuple<ll, ll, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
-using tlli = tuple<ll, ll, int>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
-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>;
@@ -211,6 +187,7 @@ template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __
 #define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
 #define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
 #define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
 template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
     return in >> p.first >> p.second;
 }
@@ -429,6 +406,23 @@ istream& operator>>(istream& in, MLLd& num) {
 }
 
 // miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
 #define functor(func) [&](auto&&... val) \
 noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
 {return func(std::forward<decltype(val)>(val)...);}
@@ -487,7 +481,7 @@ array<T, N> __initarray(const T& init) {
 }
 /////////////////////////////////////////////////////////
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -499,33 +493,36 @@ void prep() {
 }
 
 void solve() {
-    read(ll, n, k);
+    read(int, n);
-    if (k & 1) {
+    readvec(int, a, n);
-        cout << "No\n";
+    constexpr int Z = 1000;
-        return;
+    using mll = MLL<MDL>;
+    vector dp(n + 1, array<mll, 2 * Z + 1>());
+
+    mll res = 1;
+    dp[0][Z] = 1;
+    unordered_set<int> oc;
+    for (int i = 1; i <= n; ++i) {
+        for (int j = 0; j <= 2 * Z; ++j) {
+            if (j - a[i - 1] == Z) {
+                res -= dp[i - 1][j - a[i - 1]];
+                if (oc.count(a[i - 1])) {
+                    ;;
+                } else {
+                    oc.emplace(a[i - 1]);
+                    dp[i][j] += dp[i - 1][j - a[i - 1]];
                 }
-    vector<int> res(n + 1);
+            } else if (j - a[i - 1] >= 0 and j - a[i - 1] <= 2 * Z) {
-    k /= 2;
+                dp[i][j] += dp[i - 1][j - a[i - 1]];
-    iota(res.begin(), res.end(), 0);
-    int mx = n - 1;
-    int l = 1, r = n;
-    while (k) {
-        while (mx > k) {
-            l += 1;
-            mx -= 1;
             }
-        if (l >= r) {
+            dp[i][j] += dp[i - 1][j];
-            cout << "No\n";
-            return;
         }
-        k -= mx;
-        res[l] = r, res[r] = l;
-        l += 1;
-        r -= 1;
-        mx -= 2;
     }
-    cout << "Yes\n";
+
-    putvec1(res);
+    debug(accumulate(dp[n].begin(), dp[n].end(), mll(0)));
+    res += accumulate(dp[n].begin(), dp[n].end(), mll(0));
+
+    cout << res << '\n';
 }
 
 int main() {

src/bin/cf-1534e.cc

@@ -0,0 +1,566 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+int query(const vector<int>& a) {
+    if (a.empty()) return 0;
+    cout << "?";
+    for (auto&& x : a) {
+        cout << ' ' << x;
+    }
+    cout << endl;
+    read(int, x);
+    return x;
+}
+
+void solve() {
+    read(int, n, k);
+    if ((n % k) & 1) {
+        cout << -1 << endl;
+        return;
+    }
+
+    int ptr = 0;
+    int res = 0;
+    for (int i = 0; i < n / k; ++i) {
+        vector<int> curr;
+        for (int j = ptr; j < ptr + k; ++j) {
+            curr.emplace_back(j + 1);
+        }
+        ptr += k;
+        res ^= query(curr);
+    }
+
+    if (n % k) {
+        int x = (n % k) / 2;
+        vector<int> curr;
+        for (int j = n - x; j > n - x - k; --j) {
+            curr.emplace_back(j);
+        }
+        res ^= query(curr);
+        curr.clear();
+        for (int j = n; j > n - x; --j) {
+            curr.emplace_back(j);
+        }
+        for (int j = n / k * k; j > n - x - k; --j) {
+            curr.emplace_back(j);
+        }
+        res ^= query(curr);
+    }
+    cout << "! " << res << endl;
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1536e.cc

@@ -0,0 +1,536 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+constexpr int N = 4000010;
+using mll = MLL<MDL>;
+mll pw[N];
+void prep() {
+    pw[0] = 1;
+    for (int i = 1; i < N; ++i) {
+        pw[i] = pw[i - 1] * 2;
+    }
+}
+
+void solve() {
+    read(int, n, m);
+    readvec(char, a, n * m);
+    int cnt = count(a.begin(), a.end(), '0');
+    if (cnt == 0) {
+        cout << pw[n * m] - 1 << '\n';
+    } else {
+        cout << pw[n * m - cnt] << '\n';
+    }
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1537e1.cc

@@ -0,0 +1,535 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#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);
+    read(string, s);
+    string res(k, 'z');
+    for (int i = 1; i <= n; ++i) {
+        string curr;
+        int ptr = 0;
+        for (int j = 0; j < k; ++j) {
+            curr += s[ptr];
+            ptr = (ptr + 1) % i;
+        }
+        if (curr < res) res = curr;
+    }
+
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1537e2.cc

@@ -0,0 +1,555 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#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);
+    read(string, s);
+    auto z = calc_z(s);
+    int p = n;
+    int pos = INF;
+    char c = 0;
+    for (int i = 1; i <= n; ++i) {
+        int curr = i;
+        if (z[i % n] == n - i) {
+            curr += n - i;
+            if (z[n - i] != i) {
+                curr += z[n - i];
+                if (s[curr % i] < s[curr % n] and (curr < pos or curr == pos and s[pos % i] < c)) {
+                    pos = curr;
+                    c = s[pos % i];
+                    p = i;
+                }
+            }
+        } else {
+            curr += z[i % n];
+            if (s[curr % i] < s[curr % n] and (curr < pos or curr == pos and s[pos % i] < c)) {
+                pos = curr;
+                c = s[pos % i];
+                p = i;
+            }
+        }
+    }
+
+    int ptr = 0;
+    for (int i = 0; i < k; ++i) {
+        cout << s[ptr];
+        ptr = (ptr + 1) % p;
+    }
+    cout << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1538e.cc

@@ -0,0 +1,635 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 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 = hash<T>()(a.first);
+        auto hash2 = hash<U>()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+struct return_t {
+    ll length = 0;
+    string raw = "";
+    ll cnt_haha = 0;
+    int right_h = 0;
+    int right_ha = 0;
+    int right_hah = 0;
+    int left_a = 0;
+    int left_ha = 0;
+    int left_aha = 0;
+};
+
+void solve() {
+    read(int, n);
+    unordered_map<string, int> cnt;
+    unordered_map<pair<int, string>, pair<pair<int, string>, pair<int, string>>, pair_hash> var;
+    pair<int, string> last;
+    for (int i = 0; i < n; ++i) {
+        read(string, s, op);
+        if (op == ":=") {
+            read(string, v);
+            var[{cnt[s]++, s}] = { {-1, ""}, {-1, v} };
+        } else {
+            read(string, v1, plus, v2);
+            var[{cnt[s]++, s}] = { {cnt[v1] - 1, v1}, {cnt[v2] - 1, v2} };
+        }
+        if (i + 1 == n) {
+            last = { cnt[s] - 1, s };
+        }
+    }
+
+    auto make_raw = [] (const string& v) {
+        return_t res;
+        res.length = v.size();
+        if (v.size() < 3) {
+            res.raw = v;
+        }
+        for (int i = 0; i + 3 < res.length; ++i) {
+            if (v.substr(i, 4) == "haha") {
+                res.cnt_haha = 1;
+            }
+        }
+        if (v.size() >= 3 and v.substr(v.size() - 3, 3) == "hah") {
+            res.right_hah = 1;
+        }
+        if (v.size() >= 2 and v.substr(v.size() - 2, 2) == "ha") {
+            res.right_ha = 1;
+        }
+        if (v[v.size() - 1] == 'h') {
+            res.right_h = 1;
+        }
+        if (v.size() >= 3 and v.substr(0, 3) == "aha") {
+            res.left_aha = 1;
+        }
+        if (v.size() >= 2 and v.substr(0, 2) == "ha") {
+            res.left_ha = 1;
+        }
+        if (v[0] == 'a') {
+            res.left_a = 1;
+        }
+        return res;
+    };
+
+    unordered_map<pair<int, string>, return_t, pair_hash> cache;
+    auto dfs = [&] (auto dfs, const pair<int, string>& key) -> return_t {
+        return_t res;
+        if (cache.count(key)) {
+            res = cache[key];
+        } else {
+            if (var[key].first.first == -1) {
+                res = make_raw(var[key].second.second);
+            } else {
+                auto&& [l, r] = var[key];
+                return_t left_res = dfs(dfs, l), right_res = dfs(dfs, r);
+                if (left_res.length + right_res.length < 3) {
+                    res = make_raw(left_res.raw + right_res.raw);
+                } else {
+                    res.length = left_res.length + right_res.length;
+                    res.cnt_haha = left_res.cnt_haha + right_res.cnt_haha;
+                    if (left_res.right_hah and right_res.left_a) {
+                        res.cnt_haha += 1;
+                    }
+                    if (left_res.right_ha and right_res.left_ha) {
+                        res.cnt_haha += 1;
+                    }
+                    if (left_res.right_h and right_res.left_aha) {
+                        res.cnt_haha += 1;
+                    }
+                    res.right_hah = right_res.right_hah;
+                    res.right_ha = right_res.right_ha;
+                    res.right_h = right_res.right_h;
+                    res.left_aha = left_res.left_aha;
+                    res.left_ha = left_res.left_ha;
+                    res.left_a = left_res.left_a;
+                    if (right_res.raw == "ah" and left_res.right_h or right_res.raw == "h" and left_res.right_ha) {
+                        res.right_hah = 1;
+                    }
+                    if (right_res.raw == "a" and left_res.right_h) {
+                        res.right_ha = 1;
+                    }
+                    if (left_res.raw == "h" and right_res.left_a) {
+                        res.left_ha = 1;
+                    }
+                    if (left_res.raw == "a" and right_res.left_ha or left_res.raw == "ah" and right_res.left_a) {
+                        res.left_aha = 1;
+                    }
+                }
+            }
+            cache[key] = res;
+            // deb(key, cache[key].cnt_haha, cache[key].right_h, cache[key].left_aha, cache[key].length);
+        }
+        return cache[key];
+    };
+
+    cout << dfs(dfs, last).cnt_haha << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1547g.cc

@@ -0,0 +1,738 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+// #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, m);
+    adj(ch, n);
+    vector<pii> edges;
+    vector<bool> self_loop(n + 1);
+    for (int i = 0; i < m; ++i) {
+        read(int, u, v);
+        if (u != v) {
+            Edge(ch, u, v);
+            edges.emplace_back(u, v);
+        } else {
+            self_loop[u] = 1;
+        }
+    }
+    auto scc = tarjan::scc(ch);
+    int N = *max_element(scc.begin(), scc.end());
+
+    vector<int> rev(N + 1);
+    for (int i = 1; i <= n; ++i) {
+        rev[scc[i]] = i;
+    }
+    vector<int> cnt(N + 1);
+    for (int i = 1; i <= n; ++i) {
+        cnt[scc[i]] += 1;
+    }
+
+    vector<unordered_set<int, safe_hash>> nch(N + 1);
+    vector<int> ind(N + 1);
+    for (auto&& [u, v] : edges) {
+        if (scc[u] != scc[v] and not nch[scc[u]].count(scc[v])) {
+            nch[scc[u]].emplace(scc[v]);
+            ++ind[scc[v]];
+        }
+    }
+
+    deque<int> q;
+    vector<int> res(N + 1);
+    for (int i = 1; i <= N; ++i) {
+        if (not ind[i]) {
+            q.emplace_back(i);
+        }
+    }
+
+    while (q.size()) {
+        int v = q.front();  q.pop_front();
+        if (v == scc[1]) {
+            res[v] = cnt[v] > 1 or self_loop[rev[v]] ? -1 : 1;
+        } else if (res[v] and (cnt[v] > 1 or self_loop[rev[v]])) {
+            res[v] = -1;
+        }
+        for (auto&& u : nch[v]) {
+            if (res[u] == 0) {
+                if (res[v] != 0) {
+                    res[u] = res[v];
+                }
+            } else if (res[u] == 1) {
+                if (res[v] == 1 or res[v] == 2) {
+                    res[u] = 2;
+                } else if (res[v] == -1) {
+                    res[u] = -1;
+                }
+            } else if (res[u] == 2) {
+                if (res[v] == -1) {
+                    res[u] = -1;
+                }
+            }
+            if (--ind[u] == 0) {
+                q.emplace_back(u);
+            }
+        }
+    }
+
+    for (int i = 1; i <= n; ++i) {
+        cout << res[scc[i]] << " \n"[i == n];
+    }
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1550d.cc

@@ -0,0 +1,544 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+// #define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+constexpr int N = 2e5 + 10;
+using mll = MLL<MDL>;
+mll fact[N];
+void prep() {
+    fact[0] = 1;
+    for (int i = 1; i < N; ++i) {
+        fact[i] = fact[i - 1] * i;
+    }
+}
+
+void solve() {
+    read(int, n, l, r);
+
+    mll res = 0;
+    for (int x = max(1, r - n); x <= min(r - (n + 1) / 2, min(n - l, (r - n - l + 1) / 2)); ++x) {
+        res += comb(r - x, (n + 1) / 2);
+    }
+    res += min(max(0, r - n - 1), n - l) * comb(n, (n + 1) / 2);
+
+    if (n & 1) {
+        for (int x = max(1, r - n); x <= min(r - n / 2, min(n - l, (r - n - l + 1) / 2)); ++x) {
+            res += comb(r - x, n / 2);
+        }
+        res += min(max(0, r - n - 1), n - l) * comb(n, n / 2);
+    }
+
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1553f.cc

@@ -0,0 +1,661 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+template<typename Addable_Info_t, typename Tag_t, typename Sequence = std::vector<Addable_Info_t>> class segtree {
+private:
+    using size_type = uint64_t;
+    using info_type = Addable_Info_t;
+    using tag_type = Tag_t;
+    size_type _max;
+    vector<info_type> d;
+    vector<tag_type> b;
+    void pull(size_type p) {
+        d[p] = d[p * 2] + d[p * 2 + 1];
+    }
+    void push(size_type p, size_type left_len, size_type right_len) {
+        d[p * 2].apply(b[p], left_len), d[p * 2 + 1].apply(b[p], right_len);
+        b[p * 2].apply(b[p]), b[p * 2 + 1].apply(b[p]);
+        b[p] = tag_type();
+    }
+    void set(size_type s, size_type t, size_type p, size_type x, const info_type& c) {
+        if (s == t) {
+            d[p] = c;
+            return;
+        }
+        size_type m = s + (t - s >> 1);
+        if (s != t) push(p, m - s + 1, t - m);
+        if (x <= m) set(s, m, p * 2, x, c);
+        else set(m + 1, t, p * 2 + 1, x, c);
+        pull(p);
+    }
+
+    void range_apply(size_type s, size_type t, size_type p, size_type l, size_type r, const tag_type& c) {
+        if (l <= s && t <= r) {
+            d[p].apply(c, t - s + 1);
+            b[p].apply(c);
+            return;
+        }
+        size_type m = s + (t - s >> 1);
+        push(p, m - s + 1, t - m);
+        if (l <= m) range_apply(s, m, p * 2, l, r, c);
+        if (r > m)  range_apply(m + 1, t, p * 2 + 1, l, r, c);
+        pull(p);
+    }
+    info_type range_query(size_type s, size_type t, size_type p, size_type l, size_type r) {
+        if (l <= s && t <= r) {
+            return d[p];
+        }
+        size_type m = s + (t - s >> 1);
+        info_type res = {};
+        push(p, m - s + 1, t - m);
+        if (l <= m) res = res + range_query(s, m, p * 2, l, r);
+        if (r > m)  res = res + range_query(m + 1, t, p * 2 + 1, l, r);
+        return res;
+    }
+    void build(const Sequence& a, size_type s, size_type t, size_type p) {
+        if (s == t) {
+            d[p] = a[s];
+            return;
+        }
+        int m = s + (t - s >> 1);
+        build(a, s, m, p * 2);
+        build(a, m + 1, t, p * 2 + 1);
+        pull(p);
+    }
+public:
+    segtree(size_type __max) : d(4 * __max), b(4 * __max), _max(__max - 1) {}
+    segtree(const Sequence& a) : segtree(a.size()) {
+        build(a, {}, _max, 1);
+    }
+    void set(size_type i, const info_type& c) {
+        set({}, _max, 1, i, c);
+    }
+
+    void range_apply(size_type l, size_type r, const tag_type& c) {
+        range_apply({}, _max, 1, l, r, c);
+    }
+    void apply(size_type i, const tag_type& c) {
+        range_apply(i, i, c);
+    }
+    info_type range_query(size_type l, size_type r) {
+        return range_query({}, _max, 1, l, r);
+    }
+    info_type query(size_type i) {
+        return range_query(i, i);
+    }
+    Sequence serialize() {
+        Sequence res = {};
+        for (size_type i = 0; i <= _max; ++i) {
+            res.push_back(query(i));
+        }
+        return res;
+    }
+    const vector<info_type>& get_d() {
+        return d;
+    }
+};
+struct Tag {
+    ll val = 0;
+    void apply(const Tag& rhs) {
+        val += rhs.val;
+    }
+};
+struct Info {
+    ll val = 0;
+    void apply(const Tag& rhs, size_t len) {
+        val += rhs.val * len;
+    }
+};
+struct WInfo {
+    ll val = 0, cnt = 0, w = 0;
+    void apply(const Tag& rhs, size_t len) {
+        cnt += rhs.val;
+        val += w * rhs.val;
+    }
+};
+Info operator+(const Info &a, const Info &b) {
+    return {a.val + b.val};
+}
+WInfo operator+(const WInfo &a, const WInfo &b) {
+    return {a.val + b.val, a.cnt + b.cnt, 0};
+}
+
+void solve() {
+    read(int, n);
+    readvec(ll, a, n);
+
+    constexpr ll N = 3e5 + 10;
+    segtree<Info, Tag> tr(N);
+    segtree<WInfo, Tag> tr2(N);
+    for (int i = 0; i < N; ++i) {
+        tr2.set(i, { 0, 0, i });
+    }
+
+    ll res = 0;
+    for (int i = 0; i < n; ++i) {
+        res += a[i] * i - tr.query(a[i]).val;
+        for (ll j = 0; j < N; j += a[i]) {
+            tr.range_apply(j, min(N - 1, j + a[i] - 1), { j });
+            auto [sum, cnt, _] = tr2.range_query(j, min(N - 1, j + a[i] - 1));
+            res += sum - cnt * j;
+        }
+        tr2.apply(a[i], { 1 });
+        cout << res << " \n"[i + 1 == n];
+    }
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1989c.cc

@@ -0,0 +1,552 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+// #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, b, n);
+    readvec(int, c, n);
+    vector<pii> a;
+    for (int i = 0; i < n; ++i) {
+        a.emplace_back(b[i], c[i]);
+    }
+    int s1 = 0, s2 = 0, cand = 0;
+    for (int i = 0; i < n; ++i) {
+        if (a[i].first == 1 and a[i].second == 1) {
+            s1 += 1, s2 += 1, cand += 1;
+        } else if (a[i].first == 1 and a[i].second == 0) {
+            s1 += 1;
+        } else if (a[i].first == 1 and a[i].second == -1) {
+            s1 += 1;
+        } else if (a[i].first == 0 and a[i].second == 1) {
+            s2 += 1;
+        } else if (a[i].first == 0 and a[i].second == -1) {
+        } else if (a[i].first == -1 and a[i].second == 1) {
+            s2 += 1;
+        } else if (a[i].first == -1 and a[i].second == 0) {
+        } else if (a[i].first == -1 and a[i].second == -1) {
+            cand += 1;
+        }
+    }
+    int res = -INF;
+    for (int x = 0; x <= cand; ++x) {
+        int y = cand - x;
+        chmax(res, min(s1 - x, s2 - y));
+    }
+    cout << res << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1989d.cc

@@ -0,0 +1,558 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+void solve() {
+    constexpr ll N = 1e6 + 10;
+    read(int, n, m);
+    readvec(int, b, n);
+    vector<int> a(N, INF);
+    for (int i = 0; i < n; ++i) {
+        read(int, x);
+        chmin(a[b[i]], b[i] - x);
+    }
+    vector<int> ps(N, 0);
+    for (int i = 1; i < N; ++i) {
+        if (a[i] < a[ps[i - 1]]) {
+            ps[i] = i;
+        } else {
+            ps[i] = ps[i - 1];
+        }
+    }
+    vector<int> dp(N);
+    for (int i = 1; i < N; ++i) {
+        if (a[ps[i]] == INF) {
+            dp[i] = 0;
+        } else {
+            dp[i] = dp[i - a[ps[i]]] + 1;
+        }
+    }
+
+    ll res = 0;
+    while (m--) {
+        read(int, x);
+        ll curr = 0;
+        if (x >= N) {
+            curr += (x - N + 1) / a[ps[N - 1]] + 1;
+            x -= curr * a[ps[N - 1]];
+        }
+        curr += dp[x];
+        res += curr;
+    }
+    cout << res * 2 << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/cf-1989e.cc

@@ -0,0 +1,551 @@
+#pragma GCC optimize("Ofast")
+/////////////////////////////////////////////////////////
+/**
+ * This code should require C++14.
+ * However, it's only been tested with 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;
+#endif
+using int128 = __int128_t;
+using uint128 = __uint128_t;
+using ld = long double;
+using pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
+template <typename T> using max_heap = priority_queue<T>;
+template <typename T> using min_heap = priority_queue<T, vector<T>, greater<>>;
+template <typename T> using oi = ostream_iterator<T>;
+template <typename T> using ii = istream_iterator<T>;
+
+/* constants */
+constexpr int INF = 0x3f3f3f3f;
+constexpr ll INFLL = 0x3f3f3f3f3f3f3f3fLL;
+constexpr ll MDL = 1e9 + 7;
+constexpr ll PRIME = 998'244'353;
+constexpr ll MDL1 = 8784491;
+constexpr ll MDL2 = PRIME;
+constexpr int128 INT128_MAX = numeric_limits<int128>::max();
+constexpr uint128 UINT128_MAX = numeric_limits<uint128>::max();
+constexpr int128 INT128_MIN = numeric_limits<int128>::min();
+constexpr uint128 UINT128_MIN = numeric_limits<uint128>::min();
+
+/* random */
+
+mt19937 rd(chrono::duration_cast<chrono::milliseconds>(chrono::system_clock::now().time_since_epoch()).count());
+
+/* bit-wise operations */
+#define lowbit(x) ((x) & -(x))
+#define popcount(x) (__builtin_popcountll(ll(x)))
+#define parity(x) (__builtin_parityll(ll(x)))
+#define msp(x) (63LL - __builtin_clzll(ll(x)))
+#define lsp(x) (__builtin_ctzll(ll(x)))
+
+/* arithmetic operations */
+#define mod(x, y) ((((x) % (y)) + (y)) % (y))
+
+/* fast pairs */
+#define upair ull
+#define umake(x, y) (ull(x) << 32 | (ull(y) & ((1ULL << 32) - 1)))
+#define u1(p) ((p) >> 32)
+#define u2(p) ((p) & ((1ULL << 32) - 1))
+#define ult std::less<upair>
+#define ugt std::greater<upair>
+
+#define ipair ull
+#define imake(x, y) (umake(x, y))
+#define i1(p) (int(u1(ll(p))))
+#define i2(p) (ll(u2(p) << 32) >> 32)
+struct ilt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) < i2(b);
+        else return i1(a) < i1(b);
+    }
+};
+struct igt {
+    bool operator()(const ipair& a, const ipair& b) const {
+        if (i1(a) == i1(b)) return i2(a) > i2(b);
+        else return i1(a) > i1(b);
+    }
+};
+
+/* conditions */
+#define loop while (1)
+#define if_or(var, val) if (!(var == val)) var = val; else
+#define continue_or(var, val) __AS_PROCEDURE(if (var == val) continue; var = val;)
+#define break_or(var, val) __AS_PROCEDURE(if (var == val) break; var = val;)
+
+/* hash */
+struct safe_hash {
+    // https://codeforces.com/blog/entry/62393
+    static uint64_t splitmix64(uint64_t x) {
+        // http://xorshift.di.unimi.it/splitmix64.c
+        x += 0x9e3779b97f4a7c15;
+        x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9;
+        x = (x ^ (x >> 27)) * 0x94d049bb133111eb;
+        return x ^ (x >> 31);
+    }
+
+    size_t operator()(uint64_t x) const {
+        static const uint64_t FIXED_RANDOM = chrono::steady_clock::now().time_since_epoch().count();
+        return splitmix64(x + FIXED_RANDOM);
+    }
+};
+
+struct pair_hash {
+    template <typename T, typename U>
+    size_t operator()(const pair<T, U>& a) const {
+        auto hash1 = safe_hash()(a.first);
+        auto hash2 = safe_hash()(a.second);
+        if (hash1 != hash2) {
+            return hash1 ^ hash2;
+        }
+        return hash1;
+    }
+};
+
+uniform_int_distribution<mt19937::result_type> dist(PRIME);
+const size_t __array_hash_b = 31, __array_hash_mdl1 = dist(rd), __array_hash_mdl2 = dist(rd);
+struct array_hash {
+    template <typename Sequence>
+    size_t operator()(const Sequence& arr) const {
+        size_t pw1 = 1, pw2 = 1;
+        size_t res1 = 0, res2 = 0;
+        for (auto&& x : arr) {
+            res1 = (res1 + x * pw1) % __array_hash_mdl1;
+            res2 = (res2 + x * pw2) % __array_hash_mdl2;
+            pw1 = (pw1 * __array_hash_b) % __array_hash_mdl1;
+            pw2 = (pw2 * __array_hash_b) % __array_hash_mdl2;
+        }
+        return res1 + res2;
+    }
+};
+
+/* build data structures */
+#define faster(um) __AS_PROCEDURE((um).reserve(1024); (um).max_load_factor(0.25);)
+#define unordered_counter(from, to) __AS_PROCEDURE(unordered_map<__as_typeof(from), size_t, safe_hash> to; for (auto&& x : from) ++to[x];)
+#define counter(from, to, cmp) __AS_PROCEDURE(map<__as_typeof(from), size_t, cmp> to; for (auto&& x : from) ++to[x];)
+#define 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;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
+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 */
+template <typename return_t>
+return_t qpow(ll b, ll p) {
+    if (b == 0 and p != 0) return 0;
+    if (p == 0) return 1;
+    return_t half = qpow<return_t>(b, p / 2);
+    if (p % 2 == 1) return half * half * b;
+    else return half * half;
+}
+
+#define comb(n, k) ((n) < 0 or (k) < 0 or (n) < (k) ? 0 : fact[n] / fact[k] / fact[(n) - (k)])
+
+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) : mdl(other.mdl), val(other.val) {}
+    friend MLLd operator+(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val + rhs.val, lhs.mdl)); }
+    friend MLLd operator-(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val - rhs.val, lhs.mdl)); }
+    friend MLLd operator*(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * rhs.val, lhs.mdl)); }
+    friend MLLd operator/(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, mod(lhs.val * mod(inverse(rhs.val, lhs.mdl), lhs.mdl), lhs.mdl)); }
+    friend MLLd operator%(const MLLd& lhs, const MLLd& rhs) { return MLLd(lhs.mdl, 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 T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
+#define functor(func) [&](auto&&... val) \
+noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
+{return func(std::forward<decltype(val)>(val)...);}
+template <typename Func, typename RandomIt> void sort_by_key(RandomIt first, RandomIt last, Func extractor) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return std::less<>()(extractor(a), extractor(b)); });
+}
+template <typename Func, typename RandomIt, typename Compare> void sort_by_key(RandomIt first, RandomIt last, Func extractor, Compare comp) {
+    std::sort(first, last, [&] (auto&& a, auto&& b) { return comp(extractor(a), extractor(b)); });
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip(Iterator_T a_first, Iterator_T a_last, Iterator_U b_first, Iterator_U b_last) {
+    vector<pair<T, U>> res;
+    auto a_it = a_first;
+    auto b_it = b_first;
+    for (; not (a_it == a_last) and not (b_it == b_last); ++a_it, ++b_it) {
+        res.emplace_back(*a_it, *b_it);
+    }
+    return res;
+}
+template <typename T, typename U, typename Iterator_T, typename Iterator_U>
+vector<pair<T, U>> zip_n(Iterator_T a_first, Iterator_U b_first, size_t n) {
+    vector<pair<T, U>> res;
+    if (n > 0) {
+        res.emplace_back(*a_first, *b_first);
+        for (size_t i = 1; i != n; ++i) {
+            res.emplace_back(*++a_first, *++b_first);
+        }
+    }
+    return res;
+}
+template <typename T>
+class ArithmeticIterator : bidirectional_iterator_tag {
+public:
+    using difference_type = ptrdiff_t;
+    using value_type = T;
+private:
+    value_type value;
+public:
+    ArithmeticIterator(const T& value) : value(value) {}
+    value_type operator*() const { return value; }
+    ArithmeticIterator<T>& operator++() { ++value; return *this; }
+    ArithmeticIterator<T>& operator--() { --value; return *this; }
+    bool operator==(const ArithmeticIterator<T>& rhs) const { return value == rhs.value; }
+};
+template <typename T> vector<pair<int, T>> enumerate(const vector<T>& container) {
+    return zip<int, T>(ArithmeticIterator<int>(0), ArithmeticIterator<int>(INT_MAX), container.begin(), container.end());
+}
+#define initarray(init, N) (__initarray<decay<decltype(init)>::type, (N)>(init))
+template <typename T, size_t N>
+array<T, N> __initarray(const T& init) {
+    array<T, N> res;
+    for (size_t i = 0; i < N; ++i) {
+        res[i] = init;
+    }
+    return res;
+}
+/////////////////////////////////////////////////////////
+
+#define SINGLE_TEST_CASE
+// #define DUMP_TEST_CASE 7219
+// #define TOT_TEST_CASE 10000
+
+void dump() {}
+
+void dump_ignore() {}
+
+void prep() {
+}
+
+void solve() {
+    using mll = MLL<PRIME>;
+    read(int, n, k);
+    vector<mll> last(k + 1);
+    vector tot(n + 1, vector<mll>(k + 1));
+
+    tot[0][0] = 1;
+    last[0] = 1;
+
+    for (int i = 1; i <= n; ++i) {
+        vector<mll> curr(k + 1);
+        if (i - 2 == 0 or i == n) {
+            for (int j = 0; j <= k; ++j) {
+                curr[min(k, j + 1)] += tot[i - 2][j];
+            }
+        } else if (i - 3 >= 0) {
+            for (int j = 0; j <= k; ++j) {
+                curr[min(k, j + 1)] += tot[i - 3][j];
+            }
+        }
+        for (int j = 0; j <= k; ++j) {
+            curr[min(k, j + 1)] += last[j];
+        }
+
+        for (int j = 0; j <= k; ++j) {
+            tot[i][j] = tot[i - 1][j] + curr[j];
+        }
+        last = curr;
+    }
+
+    cout << last[k] << '\n';
+}
+
+int main() {
+#if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
+    assert(false && "incompatible compiler variant detected.");
+#endif
+    untie;
+    prep();
+#ifdef SINGLE_TEST_CASE
+    solve();
+#else
+    read(int, t);
+    for (int i = 0; i < t; ++i) {
+#ifdef DUMP_TEST_CASE
+        if (t != (TOT_TEST_CASE)) {
+            solve();
+        } else if (i + 1 == (DUMP_TEST_CASE)) {
+            dump();
+        } else {
+            dump_ignore();
+        }
+#else
+        solve();
+#endif
+    }
+#endif
+}

src/bin/ctext_out.cc

@@ -1,3 +1,7 @@
+#pragma GCC diagnostic ignored "-Wunused-const-variable"
+#pragma GCC diagnostic ignored "-Wreorder"
+// #pragma GCC diagnostic ignored "-Wreorder-ctor"
+#pragma GCC diagnostic ignored "-Wunknown-pragmas"
 #pragma GCC optimize("Ofast")
 /////////////////////////////////////////////////////////
 /**
@@ -492,37 +496,6 @@ void prep() {
 }
 
 void solve() {
-    read(int, n, k);
-    vector<int> a;
-    for (int i = 0; i < n; ++i) {
-        read(int, x);
-        --x;
-        a.emplace_back(x);
-    }
-
-    vector dp(n + 1, vector<int>(n + 1));
-    for (int i = 1; i <= n; ++i) {
-        // don't remove the current element
-        for (int j = 0; j <= n; ++j) {
-            dp[i][j] = dp[i - 1][j] + ((i - 1) - a[i - 1] == j);
-        }
-
-        // remove the current element
-        for (int j = 0; j < n; ++j) {
-            chmax(dp[i][j + 1], dp[i - 1][j]);
-        }
-    }
-
-    // debug(dp);
-
-    for (int i = 0; i <= n; ++i) {
-        if (dp[n][i] >= k) {
-            cout << i << '\n';
-            return;
-        }
-    }
-
-    cout << -1 << '\n';
 }
 
 int main() {

src/bin/d.cc

@@ -23,47 +23,22 @@ 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 pii = pair<int, int>;           using pil = pair<int, ll>;           using pid = pair<int, ld>;
-using pil = pair<int, ll>;
+using pli = pair<ll, int>;            using pll = pair<ll, ll>;            using pld = pair<ll, ld>;
-using pli = pair<ll, int>;
+using pdi = pair<ld, int>;            using pdl = pair<ld, ll>;            using pdd = pair<ld, ld>;
-using pll = pair<ll, ll>;
+using tiii = tuple<int, int, int>;    using tiil = tuple<int, int, ll>;    using tiid = tuple<int, int, ld>;
-using pid = pair<int, ld>;
+using tili = tuple<int, ll, int>;     using till = tuple<int, ll, ll>;     using tild = tuple<int, ll, ld>;
-using pdi = pair<ld, int>;
+using tidi = tuple<int, ld, int>;     using tidl = tuple<int, ld, ll>;     using tidd = tuple<int, ld, ld>;
-using pld = pair<ll, ld>;
+using tlii = tuple<ll, int, int>;     using tlil = tuple<ll, int, ll>;     using tlid = tuple<ll, int, ld>;
-using pdl = pair<ld, ll>;
+using tlli = tuple<ll, ll, int>;      using tlll = tuple<ll, ll, ll>;      using tlld = tuple<ll, ll, ld>;
-using pdd = pair<ld, ld>;
+using tldi = tuple<ll, ld, int>;      using tldl = tuple<ll, ld, ll>;      using tldd = tuple<ll, ld, ld>;
-using tlll = tuple<ll, ll, ll>;
+using tdii = tuple<ld, int, int>;     using tdil = tuple<ld, int, ll>;     using tdid = tuple<ld, int, ld>;
-using tlld = tuple<ll, ll, ld>;
+using tdli = tuple<ld, ll, int>;      using tdll = tuple<ld, ll, ll>;      using tdld = tuple<ld, ll, ld>;
-using tlli = tuple<ll, ll, int>;
+using tddi = tuple<ld, ld, int>;      using tddl = tuple<ld, ld, ll>;      using tddd = tuple<ld, ld, ld>;
-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>;
@@ -211,6 +186,7 @@ template<typename T, typename... U> void __read(T& x, U&... args) { cin >> x; __
 #define putvec1_eol(a) __AS_PROCEDURE(copy(a.begin() + 1, a.end(), oi<__as_typeof(a)::value_type>(cout, "\n"));)
 #define debug(x) __AS_PROCEDURE(cerr << #x" = " << (x) << endl;)
 #define debugvec(a) __AS_PROCEDURE(cerr << #a" = "; for (auto&& x : a) cerr << x << ' '; cerr << endl;)
+#define deb(...) debug(make_tuple(__VA_ARGS__))
 template<typename T, typename U> istream& operator>>(istream& in, pair<T, U>& p) {
     return in >> p.first >> p.second;
 }
@@ -429,6 +405,23 @@ istream& operator>>(istream& in, MLLd& num) {
 }
 
 // miscancellous
+template <typename T, typename U>
+bool chmax(T& lhs, const U& rhs) {
+    bool ret = lhs < rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+template <typename T, typename U>
+bool chmin(T& lhs, const U& rhs) {
+    bool ret = lhs > rhs;
+    if (ret) {
+        lhs = rhs;
+    }
+    return ret;
+}
+
 #define functor(func) [&](auto&&... val) \
 noexcept(noexcept(func(std::forward<decltype(val)>(val)...))) -> decltype(auto) \
 {return func(std::forward<decltype(val)>(val)...);}
@@ -487,7 +480,7 @@ array<T, N> __initarray(const T& init) {
 }
 /////////////////////////////////////////////////////////
 
-// #define SINGLE_TEST_CASE
+#define SINGLE_TEST_CASE
 // #define DUMP_TEST_CASE 7219
 // #define TOT_TEST_CASE 10000
 
@@ -499,32 +492,45 @@ void prep() {
 }
 
 void solve() {
-    read(int, n, c);
+    constexpr ll N = 1e6 + 10;
-    readvec(ll, a, n);
+    read(int, n, m);
-    a[0] += c;
+    readvec(int, b, n);
-    vector<ll> ps(n + 1), ss(n + 1);
+    vector<int> a(N, INF);
-    for (int i = 1; i <= n; ++i) {
-        ps[i] = max(ps[i - 1], a[i - 1]);
-    }
-    for (int i = n - 1; ~i; --i) {
-        ss[i] = max(ss[i + 1], a[i]);
-    }
-    ll left = 0;
     for (int i = 0; i < n; ++i) {
-        if (ps[i] < a[i] and ss[i + 1] <= a[i]) {
+        read(int, x);
-            cout << 0;
+        chmin(a[b[i]], b[i] - x);
+    }
+    vector<int> ps(N, 0);
+    for (int i = 1; i < N; ++i) {
+        if (a[i] < a[ps[i - 1]]) {
+            ps[i] = i;
         } else {
-            int res = i;
+            ps[i] = ps[i - 1];
-            if (ss[i + 1] > a[i] + left) {
-                res += 1;
         }
-            cout << res;
     }
-        cout << " \n"[i + 1 == n];
+    vector<int> dp(N);
-        left += a[i];
+    for (int i = 1; i < N; ++i) {
+        if (a[ps[i]] == INF) {
+            dp[i] = 0;
+        } else {
+            dp[i] = dp[i - a[ps[i]]] + 1;
         }
     }
 
+    ll res = 0;
+    while (m--) {
+        read(int, x);
+        ll curr = 0;
+        if (x >= N) {
+            curr += (x - N + 1) / a[ps[N - 1]] + 1;
+            x -= curr * a[ps[N - 1]];
+        }
+        curr += dp[x];
+        res += curr;
+    }
+    cout << res * 2 << '\n';
+}
+
 int main() {
 #if __cplusplus < 201402L or defined(_MSC_VER) and not defined(__clang__)
     assert(false && "incompatible compiler variant detected.");

src/bin/std.in

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

src/bin/template.cc

@@ -1,3 +1,6 @@
+#pragma GCC diagnostic ignored "-Wunused-const-variable"
+#pragma GCC diagnostic ignored "-Wreorder"
+#pragma GCC diagnostic ignored "-Wunknown-pragmas"
 #pragma GCC optimize("Ofast")
 /////////////////////////////////////////////////////////
 /**

src/bin/test.cc

@@ -1,36 +1,5 @@
 #include <bits/stdc++.h>
 using namespace std;
 
-class Student_Info {
-friend bool compare(Student_Info, Student_Info);
-private:
-    string name;
-    double grade;
-
-public:
-    Student_Info(string name, double grade) {
-        this->name = name;
-        this->grade = grade;
-    }
-
-    string get_name() {
-        return this->name;
-    }
-
-};
-
-bool compare(Student_Info a, Student_Info b) {
-    return a.grade < b.grade;
-}
-
 int main() {
-    vector<Student_Info> students = {
-        Student_Info("John", 8.5),
-        Student_Info("Bob", 7.5),
-        Student_Info("Jane", 9.5),
-    };
-    sort(students.begin(), students.end(), compare);
-    for (auto&& student : students) {
-        cout << student.get_name() << endl;
-    }
 }

src/bin/test.py

@@ -7,16 +7,32 @@ import io
 PRIME = 998_244_353
 
 if __name__ == '__main__':
-    a = [(0, 0), (2, 16), (30, 14), (4, 6), (2, 10)]
+    n = 4
-    a.sort(key=lambda t: t[1])
+    k = 5
-    n = len(a)
+    for i in range(1 << n * k):
-    res = []
+        j = i
-    for i in range(n):
+        cnt = (j & ((1 << n) - 1)).bit_count()
-        for j in range(i + 1, n):
+        f = 1
-            for k in range(i + 1, j):
+        for _ in range(k):
-                # if (a[j][0] - a[i][0]) * (a[k][1] - a[i][1]) % (a[j][1] - a[i][1]):
+            if (j & ((1 << n) - 1)).bit_count() != cnt:
-                #     continue
+                f = 0
-                x, y = a[i][0] // 2 + (a[j][0] // 2 - a[i][0] // 2) * (a[k][1] // 2 - a[i][1] // 2 + a[j][1] // 2 - a[i][1] // 2 - 1) // (a[j][1] // 2 - a[i][1] // 2) - a[k][0] // 2, a[i][1] // 2 - a[j][1] // 2
+                break
-                if (x + y) & 1:
+            j >>= n
-                    res.append((a[i], a[j], a[k], x, y))
+        for j in range(n):
-    print(res)
+            l = i >> j
+            c = 0
+            while l:
+                c += l & 1
+                l >>= n
+            if (c & 1) == 0:
+                f = 0
+                break
+        if f:
+            m = i
+            for j in range(k):
+                for l in range(n):
+                    print(m & 1, end='')
+                    m >>= 1
+                print()
+            print()
+