結果
問題 | No.2798 Multiple Chain |
ユーザー | shiomusubi496 |
提出日時 | 2024-06-28 22:40:49 |
言語 | C++17 (gcc 12.3.0 + boost 1.83.0) |
結果 |
AC
|
実行時間 | 384 ms / 2,000 ms |
コード長 | 45,281 bytes |
コンパイル時間 | 3,215 ms |
コンパイル使用メモリ | 245,340 KB |
実行使用メモリ | 5,864 KB |
最終ジャッジ日時 | 2024-06-28 22:40:55 |
合計ジャッジ時間 | 6,102 ms |
ジャッジサーバーID (参考情報) |
judge3 / judge4 |
(要ログイン)
テストケース
テストケース表示入力 | 結果 | 実行時間 実行使用メモリ |
---|---|---|
testcase_00 | AC | 2 ms
5,248 KB |
testcase_01 | AC | 2 ms
5,376 KB |
testcase_02 | AC | 8 ms
5,376 KB |
testcase_03 | AC | 2 ms
5,376 KB |
testcase_04 | AC | 2 ms
5,376 KB |
testcase_05 | AC | 55 ms
5,376 KB |
testcase_06 | AC | 39 ms
5,376 KB |
testcase_07 | AC | 2 ms
5,376 KB |
testcase_08 | AC | 2 ms
5,376 KB |
testcase_09 | AC | 2 ms
5,376 KB |
testcase_10 | AC | 2 ms
5,376 KB |
testcase_11 | AC | 2 ms
5,376 KB |
testcase_12 | AC | 2 ms
5,376 KB |
testcase_13 | AC | 4 ms
5,376 KB |
testcase_14 | AC | 10 ms
5,376 KB |
testcase_15 | AC | 39 ms
5,376 KB |
testcase_16 | AC | 47 ms
5,608 KB |
testcase_17 | AC | 51 ms
5,864 KB |
testcase_18 | AC | 50 ms
5,604 KB |
testcase_19 | AC | 57 ms
5,732 KB |
testcase_20 | AC | 2 ms
5,376 KB |
testcase_21 | AC | 2 ms
5,376 KB |
testcase_22 | AC | 3 ms
5,376 KB |
testcase_23 | AC | 2 ms
5,376 KB |
testcase_24 | AC | 2 ms
5,376 KB |
testcase_25 | AC | 7 ms
5,376 KB |
testcase_26 | AC | 6 ms
5,376 KB |
testcase_27 | AC | 4 ms
5,376 KB |
testcase_28 | AC | 6 ms
5,376 KB |
testcase_29 | AC | 4 ms
5,376 KB |
testcase_30 | AC | 10 ms
5,376 KB |
testcase_31 | AC | 6 ms
5,376 KB |
testcase_32 | AC | 13 ms
5,376 KB |
testcase_33 | AC | 7 ms
5,376 KB |
testcase_34 | AC | 27 ms
5,376 KB |
testcase_35 | AC | 7 ms
5,376 KB |
testcase_36 | AC | 8 ms
5,376 KB |
testcase_37 | AC | 12 ms
5,376 KB |
testcase_38 | AC | 8 ms
5,376 KB |
testcase_39 | AC | 8 ms
5,376 KB |
testcase_40 | AC | 110 ms
5,376 KB |
testcase_41 | AC | 223 ms
5,376 KB |
testcase_42 | AC | 384 ms
5,376 KB |
testcase_43 | AC | 223 ms
5,376 KB |
testcase_44 | AC | 153 ms
5,376 KB |
testcase_45 | AC | 2 ms
5,376 KB |
testcase_46 | AC | 2 ms
5,376 KB |
testcase_47 | AC | 2 ms
5,376 KB |
testcase_48 | AC | 3 ms
5,376 KB |
testcase_49 | AC | 3 ms
5,376 KB |
testcase_50 | AC | 3 ms
5,376 KB |
testcase_51 | AC | 3 ms
5,376 KB |
testcase_52 | AC | 9 ms
5,376 KB |
testcase_53 | AC | 6 ms
5,376 KB |
ソースコード
#line 2 "library/other/template.hpp" #include <bits/stdc++.h> #line 2 "library/template/macros.hpp" #line 4 "library/template/macros.hpp" #ifndef __COUNTER__ #define __COUNTER__ __LINE__ #endif #define OVERLOAD5(a, b, c, d, e, ...) e #define REP1_0(b, c) REP1_1(b, c) #define REP1_1(b, c) \ for (ll REP_COUNTER_##c = 0; REP_COUNTER_##c < (ll)(b); ++REP_COUNTER_##c) #define REP1(b) REP1_0(b, __COUNTER__) #define REP2(i, b) for (ll i = 0; i < (ll)(b); ++i) #define REP3(i, a, b) for (ll i = (ll)(a); i < (ll)(b); ++i) #define REP4(i, a, b, c) for (ll i = (ll)(a); i < (ll)(b); i += (ll)(c)) #define rep(...) OVERLOAD5(__VA_ARGS__, REP4, REP3, REP2, REP1)(__VA_ARGS__) #define RREP2(i, a) for (ll i = (ll)(a)-1; i >= 0; --i) #define RREP3(i, a, b) for (ll i = (ll)(a)-1; i >= (ll)(b); --i) #define RREP4(i, a, b, c) for (ll i = (ll)(a)-1; i >= (ll)(b); i -= (ll)(c)) #define rrep(...) OVERLOAD5(__VA_ARGS__, RREP4, RREP3, RREP2)(__VA_ARGS__) #define REPS2(i, b) for (ll i = 1; i <= (ll)(b); ++i) #define REPS3(i, a, b) for (ll i = (ll)(a) + 1; i <= (ll)(b); ++i) #define REPS4(i, a, b, c) for (ll i = (ll)(a) + 1; i <= (ll)(b); i += (ll)(c)) #define reps(...) OVERLOAD5(__VA_ARGS__, REPS4, REPS3, REPS2)(__VA_ARGS__) #define RREPS2(i, a) for (ll i = (ll)(a); i > 0; --i) #define RREPS3(i, a, b) for (ll i = (ll)(a); i > (ll)(b); --i) #define RREPS4(i, a, b, c) for (ll i = (ll)(a); i > (ll)(b); i -= (ll)(c)) #define rreps(...) OVERLOAD5(__VA_ARGS__, RREPS4, RREPS3, RREPS2)(__VA_ARGS__) #define each_for(...) for (auto&& __VA_ARGS__) #define each_const(...) for (const auto& __VA_ARGS__) #define all(v) std::begin(v), std::end(v) #define rall(v) std::rbegin(v), std::rend(v) #if __cpp_if_constexpr >= 201606L #define IF_CONSTEXPR constexpr #else #define IF_CONSTEXPR #endif #define IO_BUFFER_SIZE (1 << 17) #line 2 "library/template/alias.hpp" #line 4 "library/template/alias.hpp" using ll = long long; using uint = unsigned int; using ull = unsigned long long; using i128 = __int128_t; using u128 = __uint128_t; using ld = long double; using PLL = std::pair<ll, ll>; template<class T> using prique = std::priority_queue<T, std::vector<T>, std::greater<T>>; template<class T> struct infinity { static constexpr T value = std::numeric_limits<T>::max() / 2; static constexpr T mvalue = std::numeric_limits<T>::lowest() / 2; static constexpr T max = std::numeric_limits<T>::max(); static constexpr T min = std::numeric_limits<T>::lowest(); }; #if __cplusplus <= 201402L template<class T> constexpr T infinity<T>::value; template<class T> constexpr T infinity<T>::mvalue; template<class T> constexpr T infinity<T>::max; template<class T> constexpr T infinity<T>::min; #endif #if __cpp_variable_templates >= 201304L template<class T> constexpr T INF = infinity<T>::value; #endif constexpr ll inf = infinity<ll>::value; constexpr ld EPS = 1e-8; constexpr ld PI = 3.1415926535897932384626; #line 2 "library/template/type_traits.hpp" #line 5 "library/template/type_traits.hpp" template<class T, class... Args> struct function_traits_impl { using result_type = T; template<std::size_t idx> using argument_type = typename std::tuple_element<idx, std::tuple<Args...>>::type; using argument_tuple = std::tuple<Args...>; static constexpr std::size_t arg_size() { return sizeof...(Args); } }; template<class> struct function_traits_helper; template<class Res, class Tp, class... Args> struct function_traits_helper<Res (Tp::*)(Args...)> { using type = function_traits_impl<Res, Args...>; }; template<class Res, class Tp, class... Args> struct function_traits_helper<Res (Tp::*)(Args...)&> { using type = function_traits_impl<Res, Args...>; }; template<class Res, class Tp, class... Args> struct function_traits_helper<Res (Tp::*)(Args...) const> { using type = function_traits_impl<Res, Args...>; }; template<class Res, class Tp, class... Args> struct function_traits_helper<Res (Tp::*)(Args...) const&> { using type = function_traits_impl<Res, Args...>; }; #if __cpp_noexcept_function_type >= 201510L template<class Res, class Tp, class... Args> struct function_traits_helper<Res (Tp::*)(Args...) noexcept> { using type = function_traits_impl<Res, Args...>; }; template<class Res, class Tp, class... Args> struct function_traits_helper<Res (Tp::*)(Args...)& noexcept> { using type = function_traits_impl<Res, Args...>; }; template<class Res, class Tp, class... Args> struct function_traits_helper<Res (Tp::*)(Args...) const noexcept> { using type = function_traits_impl<Res, Args...>; }; template<class Res, class Tp, class... Args> struct function_traits_helper<Res (Tp::*)(Args...) const& noexcept> { using type = function_traits_impl<Res, Args...>; }; #endif template<class F> using function_traits = typename function_traits_helper< decltype(&std::remove_reference<F>::type::operator())>::type; template<class F> using function_result_type = typename function_traits<F>::result_type; template<class F, std::size_t idx> using function_argument_type = typename function_traits<F>::template argument_type<idx>; template<class F> using function_argument_tuple = typename function_traits<F>::argument_tuple; template<class T> using is_signed_int = std::integral_constant<bool, (std::is_integral<T>::value && std::is_signed<T>::value) || std::is_same<T, i128>::value>; template<class T> using is_unsigned_int = std::integral_constant<bool, (std::is_integral<T>::value && std::is_unsigned<T>::value) || std::is_same<T, u128>::value>; template<class T> using is_int = std::integral_constant<bool, is_signed_int<T>::value || is_unsigned_int<T>::value>; template<class T> using make_signed_int = typename std::conditional< std::is_same<T, i128>::value || std::is_same<T, u128>::value, std::common_type<i128>, std::make_signed<T>>::type; template<class T> using make_unsigned_int = typename std::conditional< std::is_same<T, i128>::value || std::is_same<T, u128>::value, std::common_type<u128>, std::make_unsigned<T>>::type; template<class T, class = void> struct is_range : std::false_type {}; template<class T> struct is_range< T, decltype(all(std::declval<typename std::add_lvalue_reference<T>::type>()), (void)0)> : std::true_type {}; template<class T, bool = is_range<T>::value> struct range_rank : std::integral_constant<std::size_t, 0> {}; template<class T> struct range_rank<T, true> : std::integral_constant<std::size_t, range_rank<typename T::value_type>::value + 1> {}; template<std::size_t size> struct int_least { static_assert(size <= 128, "size must be less than or equal to 128"); using type = typename std::conditional< size <= 8, std::int_least8_t, typename std::conditional< size <= 16, std::int_least16_t, typename std::conditional< size <= 32, std::int_least32_t, typename std::conditional<size <= 64, std::int_least64_t, i128>::type>::type>::type>::type; }; template<std::size_t size> using int_least_t = typename int_least<size>::type; template<std::size_t size> struct uint_least { static_assert(size <= 128, "size must be less than or equal to 128"); using type = typename std::conditional< size <= 8, std::uint_least8_t, typename std::conditional< size <= 16, std::uint_least16_t, typename std::conditional< size <= 32, std::uint_least32_t, typename std::conditional<size <= 64, std::uint_least64_t, u128>::type>::type>::type>::type; }; template<std::size_t size> using uint_least_t = typename uint_least<size>::type; template<class T> using double_size_int = int_least<std::numeric_limits<T>::digits * 2 + 1>; template<class T> using double_size_int_t = typename double_size_int<T>::type; template<class T> using double_size_uint = uint_least<std::numeric_limits<T>::digits * 2>; template<class T> using double_size_uint_t = typename double_size_uint<T>::type; template<class T> using double_size = typename std::conditional<is_signed_int<T>::value, double_size_int<T>, double_size_uint<T>>::type; template<class T> using double_size_t = typename double_size<T>::type; #line 2 "library/template/in.hpp" #line 4 "library/template/in.hpp" #include <unistd.h> #line 8 "library/template/in.hpp" template<std::size_t buf_size = IO_BUFFER_SIZE, std::size_t decimal_precision = 16> class Scanner { private: template<class, class = void> struct has_scan : std::false_type {}; template<class T> struct has_scan< T, decltype(std::declval<T>().scan(std::declval<Scanner&>()), (void)0)> : std::true_type {}; int fd; int idx, sz; bool state; std::array<char, IO_BUFFER_SIZE + 1> buffer; inline char cur() { if (idx == sz) load(); if (idx == sz) { state = false; return '\0'; } return buffer[idx]; } inline void next() { if (idx == sz) load(); if (idx == sz) return; ++idx; } public: inline void load() { int len = sz - idx; if (idx < len) return; std::memcpy(buffer.begin(), buffer.begin() + idx, len); sz = len + read(fd, buffer.data() + len, buf_size - len); buffer[sz] = 0; idx = 0; } Scanner(int fd) : fd(fd), idx(0), sz(0), state(true) {} Scanner(FILE* fp) : fd(fileno(fp)), idx(0), sz(0), state(true) {} inline char scan_char() { if (idx == sz) load(); return idx == sz ? '\0' : buffer[idx++]; } Scanner ignore(int n = 1) { if (idx + n > sz) load(); idx += n; return *this; } inline void discard_space() { if (idx == sz) load(); while (('\t' <= buffer[idx] && buffer[idx] <= '\r') || buffer[idx] == ' ') { if (++idx == sz) load(); } } void scan(char& a) { discard_space(); a = scan_char(); } void scan(bool& a) { discard_space(); a = scan_char() != '0'; } void scan(std::string& a) { discard_space(); a.clear(); while (cur() != '\0' && (buffer[idx] < '\t' || '\r' < buffer[idx]) && buffer[idx] != ' ') { a += scan_char(); } } template<std::size_t len> void scan(std::bitset<len>& a) { discard_space(); if (idx + len > sz) load(); rrep (i, len) a[i] = buffer[idx++] != '0'; } template<class T, typename std::enable_if<is_signed_int<T>::value && !has_scan<T>::value>::type* = nullptr> void scan(T& a) { discard_space(); if (buffer[idx] == '-') { ++idx; if (idx + 40 > sz && (idx == sz || ('0' <= buffer[sz - 1] && buffer[sz - 1] <= '9'))) load(); a = 0; while ('0' <= buffer[idx] && buffer[idx] <= '9') { a = a * 10 - (buffer[idx++] - '0'); } } else { if (idx + 40 > sz && '0' <= buffer[sz - 1] && buffer[sz - 1] <= '9') load(); a = 0; while ('0' <= buffer[idx] && buffer[idx] <= '9') { a = a * 10 + (buffer[idx++] - '0'); } } } template<class T, typename std::enable_if<is_unsigned_int<T>::value && !has_scan<T>::value>::type* = nullptr> void scan(T& a) { discard_space(); if (idx + 40 > sz && '0' <= buffer[sz - 1] && buffer[sz - 1] <= '9') load(); a = 0; while ('0' <= buffer[idx] && buffer[idx] <= '9') { a = a * 10 + (buffer[idx++] - '0'); } } template<class T, typename std::enable_if<std::is_floating_point<T>::value && !has_scan<T>::value>::type* = nullptr> void scan(T& a) { discard_space(); bool sgn = false; if (cur() == '-') { sgn = true; next(); } a = 0; while ('0' <= cur() && cur() <= '9') { a = a * 10 + cur() - '0'; next(); } if (cur() == '.') { next(); T n = 0, d = 1; for (int i = 0; '0' <= cur() && cur() <= '9' && i < (int)decimal_precision; ++i) { n = n * 10 + cur() - '0'; d *= 10; next(); } while ('0' <= cur() && cur() <= '9') next(); a += n / d; } if (sgn) a = -a; } private: template<std::size_t i, class... Args> void scan(std::tuple<Args...>& a) { if IF_CONSTEXPR (i < sizeof...(Args)) { scan(std::get<i>(a)); scan<i + 1, Args...>(a); } } public: template<class... Args> void scan(std::tuple<Args...>& a) { scan<0, Args...>(a); } template<class T, class U> void scan(std::pair<T, U>& a) { scan(a.first); scan(a.second); } template<class T, typename std::enable_if<is_range<T>::value && !has_scan<T>::value>::type* = nullptr> void scan(T& a) { for (auto&& i : a) scan(i); } template<class T, typename std::enable_if<has_scan<T>::value>::type* = nullptr> void scan(T& a) { a.scan(*this); } void operator()() {} template<class Head, class... Args> void operator()(Head& head, Args&... args) { scan(head); operator()(args...); } template<class T> Scanner& operator>>(T& a) { scan(a); return *this; } explicit operator bool() const { return state; } friend Scanner& getline(Scanner& scan, std::string& a) { a.erase(); char c; if ((c = scan.scan_char()) == '\n' || c == '\0') return scan; a += c; while ((c = scan.scan_char()) != '\n' && c != '\0') a += c; scan.state = true; return scan; } }; Scanner<> scan(0); #line 2 "library/template/out.hpp" #line 8 "library/template/out.hpp" struct NumberToString { char buf[10000][4]; constexpr NumberToString() : buf{} { rep (i, 10000) { int n = i; rrep (j, 4) { buf[i][j] = (char)('0' + n % 10); n /= 10; } } } } constexpr precalc_number_to_string; template<std::size_t buf_size = IO_BUFFER_SIZE, bool debug = false> class Printer { private: template<class, bool = debug, class = void> struct has_print : std::false_type {}; template<class T> struct has_print<T, false, decltype(std::declval<T>().print(std::declval<Printer&>()), (void)0)> : std::true_type {}; template<class T> struct has_print<T, true, decltype(std::declval<T>().debug(std::declval<Printer&>()), (void)0)> : std::true_type {}; int fd; std::size_t idx; std::array<char, buf_size> buffer; std::size_t decimal_precision; public: inline void print_char(char c) { #if SHIO_LOCAL buffer[idx++] = c; if (idx == buf_size) flush(); #else if IF_CONSTEXPR (!debug) { buffer[idx++] = c; if (idx == buf_size) flush(); } #endif } inline void flush() { idx = write(fd, buffer.begin(), idx); idx = 0; } Printer(int fd) : fd(fd), idx(0), decimal_precision(16) {} Printer(FILE* fp) : fd(fileno(fp)), idx(0), decimal_precision(16) {} ~Printer() { flush(); } void set_decimal_precision(std::size_t decimal_precision) { this->decimal_precision = decimal_precision; } void print(char c) { if IF_CONSTEXPR (debug) print_char('\''); print_char(c); if IF_CONSTEXPR (debug) print_char('\''); } void print(bool b) { print_char((char)(b + '0')); } void print(const char* a) { if IF_CONSTEXPR (debug) print_char('"'); for (; *a != '\0'; ++a) print_char(*a); if IF_CONSTEXPR (debug) print_char('"'); } template<std::size_t len> void print(const char (&a)[len]) { if IF_CONSTEXPR (debug) print_char('"'); for (auto i : a) print_char(i); if IF_CONSTEXPR (debug) print_char('"'); } void print(const std::string& a) { if IF_CONSTEXPR (debug) print_char('"'); for (auto i : a) print_char(i); if IF_CONSTEXPR (debug) print_char('"'); } template<std::size_t len> void print(const std::bitset<len>& a) { rrep (i, len) print_char((char)(a[i] + '0')); } template<class T, typename std::enable_if<is_int<T>::value && !has_print<T>::value>::type* = nullptr> void print(T a) { if (!a) { print_char('0'); return; } if IF_CONSTEXPR (is_signed_int<T>::value) { if (a < 0) { print_char('-'); using U = typename make_unsigned_int<T>::type; print(static_cast<U>(-static_cast<U>(a))); return; } } if (idx + 40 >= buf_size) flush(); static char s[40]; int t = 40; while (a >= 10000) { int i = a % 10000; a /= 10000; t -= 4; std::memcpy(s + t, precalc_number_to_string.buf[i], 4); } if (a >= 1000) { std::memcpy(buffer.begin() + idx, precalc_number_to_string.buf[a], 4); idx += 4; } else if (a >= 100) { std::memcpy(buffer.begin() + idx, precalc_number_to_string.buf[a] + 1, 3); idx += 3; } else if (a >= 10) { std::memcpy(buffer.begin() + idx, precalc_number_to_string.buf[a] + 2, 2); idx += 2; } else { buffer[idx++] = '0' | a; } std::memcpy(buffer.begin() + idx, s + t, 40 - t); idx += 40 - t; } template<class T, typename std::enable_if<std::is_floating_point<T>::value && !has_print<T>::value>::type* = nullptr> void print(T a) { if (a == std::numeric_limits<T>::infinity()) { print("inf"); return; } if (a == -std::numeric_limits<T>::infinity()) { print("-inf"); return; } if (std::isnan(a)) { print("nan"); return; } if (a < 0) { print_char('-'); a = -a; } T b = a; if (b < 1) { print_char('0'); } else { std::string s; while (b >= 1) { s += (char)('0' + (int)std::fmod(b, 10.0)); b /= 10; } for (auto i = s.rbegin(); i != s.rend(); ++i) print_char(*i); } print_char('.'); rep (decimal_precision) { a *= 10; print_char((char)('0' + (int)std::fmod(a, 10.0))); } } private: template<std::size_t i, class... Args> void print(const std::tuple<Args...>& a) { if IF_CONSTEXPR (i < sizeof...(Args)) { if IF_CONSTEXPR (debug) print_char(','); print_char(' '); print(std::get<i>(a)); print<i + 1, Args...>(a); } } public: template<class... Args> void print(const std::tuple<Args...>& a) { if IF_CONSTEXPR (debug) print_char('('); if IF_CONSTEXPR (sizeof...(Args) != 0) print(std::get<0>(a)); print<1, Args...>(a); if IF_CONSTEXPR (debug) print_char(')'); } template<class T, class U> void print(const std::pair<T, U>& a) { if IF_CONSTEXPR (debug) print_char('('); print(a.first); if IF_CONSTEXPR (debug) print_char(','); print_char(' '); print(a.second); if IF_CONSTEXPR (debug) print_char(')'); } template<class T, typename std::enable_if<is_range<T>::value && !has_print<T>::value>::type* = nullptr> void print(const T& a) { if IF_CONSTEXPR (debug) print_char('{'); for (auto i = std::begin(a); i != std::end(a); ++i) { if (i != std::begin(a)) { if IF_CONSTEXPR (debug) print_char(','); print_char(' '); } print(*i); } if IF_CONSTEXPR (debug) print_char('}'); } template<class T, typename std::enable_if<has_print<T>::value && !debug>::type* = nullptr> void print(const T& a) { a.print(*this); } template<class T, typename std::enable_if<has_print<T>::value && debug>::type* = nullptr> void print(const T& a) { a.debug(*this); } void operator()() {} template<class Head, class... Args> void operator()(const Head& head, const Args&... args) { print(head); operator()(args...); } template<class T> Printer& operator<<(const T& a) { print(a); return *this; } Printer& operator<<(Printer& (*pf)(Printer&)) { return pf(*this); } }; template<std::size_t buf_size, bool debug> Printer<buf_size, debug>& endl(Printer<buf_size, debug>& pr) { pr.print_char('\n'); pr.flush(); return pr; } template<std::size_t buf_size, bool debug> Printer<buf_size, debug>& flush(Printer<buf_size, debug>& pr) { pr.flush(); return pr; } struct SetPrec { int n; template<class Pr> void print(Pr& pr) const { pr.set_decimal_precision(n); } template<class Pr> void debug(Pr& pr) const { pr.set_decimal_precision(n); } }; SetPrec setprec(int n) { return SetPrec{n}; }; Printer<> print(1), eprint(2); void prints() { print.print_char('\n'); } template<class T> auto prints(const T& v) -> decltype(print << v, (void)0) { print << v; print.print_char('\n'); } template<class Head, class... Tail> auto prints(const Head& head, const Tail&... tail) -> decltype(print << head, (void)0) { print << head; print.print_char(' '); prints(tail...); } Printer<IO_BUFFER_SIZE, true> debug(1), edebug(2); void debugs() { debug.print_char('\n'); } template<class T> auto debugs(const T& v) -> decltype(debug << v, (void)0) { debug << v; debug.print_char('\n'); } template<class Head, class... Tail> auto debugs(const Head& head, const Tail&... tail) -> decltype(debug << head, (void)0) { debug << head; debug.print_char(' '); debugs(tail...); } #line 2 "library/template/bitop.hpp" #line 6 "library/template/bitop.hpp" namespace bitop { #define KTH_BIT(b, k) (((b) >> (k)) & 1) #define POW2(k) (1ull << (k)) inline ull next_combination(int n, ull x) { if (n == 0) return 1; ull a = x & -x; ull b = x + a; return (x & ~b) / a >> 1 | b; } #define rep_comb(i, n, k) \ for (ull i = (1ull << (k)) - 1; i < (1ull << (n)); \ i = bitop::next_combination((n), i)) inline constexpr int msb(ull x) { int res = x ? 0 : -1; if (x & 0xFFFFFFFF00000000) x &= 0xFFFFFFFF00000000, res += 32; if (x & 0xFFFF0000FFFF0000) x &= 0xFFFF0000FFFF0000, res += 16; if (x & 0xFF00FF00FF00FF00) x &= 0xFF00FF00FF00FF00, res += 8; if (x & 0xF0F0F0F0F0F0F0F0) x &= 0xF0F0F0F0F0F0F0F0, res += 4; if (x & 0xCCCCCCCCCCCCCCCC) x &= 0xCCCCCCCCCCCCCCCC, res += 2; return res + ((x & 0xAAAAAAAAAAAAAAAA) ? 1 : 0); } inline constexpr int ceil_log2(ull x) { return x ? msb(x - 1) + 1 : 0; } inline constexpr ull reverse(ull x) { x = ((x & 0xAAAAAAAAAAAAAAAA) >> 1) | ((x & 0x5555555555555555) << 1); x = ((x & 0xCCCCCCCCCCCCCCCC) >> 2) | ((x & 0x3333333333333333) << 2); x = ((x & 0xF0F0F0F0F0F0F0F0) >> 4) | ((x & 0x0F0F0F0F0F0F0F0F) << 4); x = ((x & 0xFF00FF00FF00FF00) >> 8) | ((x & 0x00FF00FF00FF00FF) << 8); x = ((x & 0xFFFF0000FFFF0000) >> 16) | ((x & 0x0000FFFF0000FFFF) << 16); return (x >> 32) | (x << 32); } inline constexpr ull reverse(ull x, int n) { return reverse(x) >> (64 - n); } } // namespace bitop inline constexpr int popcnt(ull x) noexcept { #if __cplusplus >= 202002L return std::popcount(x); #endif x = (x & 0x5555555555555555) + ((x >> 1) & 0x5555555555555555); x = (x & 0x3333333333333333) + ((x >> 2) & 0x3333333333333333); x = (x & 0x0f0f0f0f0f0f0f0f) + ((x >> 4) & 0x0f0f0f0f0f0f0f0f); x = (x & 0x00ff00ff00ff00ff) + ((x >> 8) & 0x00ff00ff00ff00ff); x = (x & 0x0000ffff0000ffff) + ((x >> 16) & 0x0000ffff0000ffff); return (x & 0x00000000ffffffff) + ((x >> 32) & 0x00000000ffffffff); } #line 2 "library/template/func.hpp" #line 6 "library/template/func.hpp" template<class T, class U, class Comp = std::less<>> inline constexpr bool chmin(T& a, const U& b, Comp cmp = Comp()) noexcept(noexcept(cmp(b, a))) { return cmp(b, a) ? a = b, true : false; } template<class T, class U, class Comp = std::less<>> inline constexpr bool chmax(T& a, const U& b, Comp cmp = Comp()) noexcept(noexcept(cmp(a, b))) { return cmp(a, b) ? a = b, true : false; } inline constexpr ll gcd(ll a, ll b) { if (a < 0) a = -a; if (b < 0) b = -b; while (b) { const ll c = a; a = b; b = c % b; } return a; } inline constexpr ll lcm(ll a, ll b) { return a / gcd(a, b) * b; } inline constexpr bool is_prime(ll N) { if (N <= 1) return false; for (ll i = 2; i * i <= N; ++i) { if (N % i == 0) return false; } return true; } inline std::vector<ll> prime_factor(ll N) { std::vector<ll> res; for (ll i = 2; i * i <= N; ++i) { while (N % i == 0) { res.push_back(i); N /= i; } } if (N != 1) res.push_back(N); return res; } inline constexpr ll my_pow(ll a, ll b) { ll res = 1; while (b) { if (b & 1) res *= a; b >>= 1; a *= a; } return res; } inline constexpr ll mod_pow(ll a, ll b, ll mod) { assert(mod > 0); if (mod == 1) return 0; a %= mod; ll res = 1; while (b) { if (b & 1) (res *= a) %= mod; b >>= 1; (a *= a) %= mod; } return res; } inline PLL extGCD(ll a, ll b) { const ll n = a, m = b; ll x = 1, y = 0, u = 0, v = 1; ll t; while (b) { t = a / b; std::swap(a -= t * b, b); std::swap(x -= t * u, u); std::swap(y -= t * v, v); } if (x < 0) { x += m; y -= n; } return {x, y}; } inline ll mod_inv(ll a, ll mod) { ll b = mod; ll x = 1, u = 0; ll t; while (b) { t = a / b; std::swap(a -= t * b, b); std::swap(x -= t * u, u); } if (x < 0) x += mod; assert(a == 1); return x; } #line 2 "library/template/util.hpp" #line 6 "library/template/util.hpp" template<class F> class RecLambda { private: F f; public: explicit constexpr RecLambda(F&& f_) : f(std::forward<F>(f_)) {} template<class... Args> constexpr auto operator()(Args&&... args) -> decltype(f(*this, std::forward<Args>(args)...)) { return f(*this, std::forward<Args>(args)...); } }; template<class F> inline constexpr RecLambda<F> rec_lambda(F&& f) { return RecLambda<F>(std::forward<F>(f)); } template<class Head, class... Tail> struct multi_dim_vector { using type = std::vector<typename multi_dim_vector<Tail...>::type>; }; template<class T> struct multi_dim_vector<T> { using type = T; }; template<class T, class Arg> constexpr std::vector<T> make_vec(int n, Arg&& arg) { return std::vector<T>(n, std::forward<Arg>(arg)); } template<class T, class... Args> constexpr typename multi_dim_vector<Args..., T>::type make_vec(int n, Args&&... args) { return typename multi_dim_vector<Args..., T>::type( n, make_vec<T>(std::forward<Args>(args)...)); } template<class T, class Comp = std::less<T>> class compressor { private: std::vector<T> dat; Comp cmp; bool sorted = false; public: compressor() : compressor(Comp()) {} compressor(const Comp& cmp) : cmp(cmp) {} compressor(const std::vector<T>& vec, bool f = false, const Comp& cmp = Comp()) : dat(vec), cmp(cmp) { if (f) build(); } compressor(std::vector<T>&& vec, bool f = false, const Comp& cmp = Comp()) : dat(std::move(vec)), cmp(cmp) { if (f) build(); } compressor(std::initializer_list<T> il, bool f = false, const Comp& cmp = Comp()) : dat(all(il)), cmp(cmp) { if (f) build(); } void reserve(int n) { assert(!sorted); dat.reserve(n); } void push_back(const T& v) { assert(!sorted); dat.push_back(v); } void push_back(T&& v) { assert(!sorted); dat.push_back(std::move(v)); } template<class... Args> void emplace_back(Args&&... args) { assert(!sorted); dat.emplace_back(std::forward<Args>(args)...); } void push(const std::vector<T>& vec) { assert(!sorted); const int n = dat.size(); dat.resize(n + vec.size()); rep (i, vec.size()) dat[n + i] = vec[i]; } int build() { assert(!sorted); sorted = true; std::sort(all(dat), cmp); dat.erase(std::unique(all(dat), [&](const T& a, const T& b) -> bool { return !cmp(a, b) && !cmp(b, a); }), dat.end()); return dat.size(); } const T& operator[](int k) const& { assert(sorted); assert(0 <= k && k < (int)dat.size()); return dat[k]; } int get(const T& val) const { assert(sorted); auto itr = std::lower_bound(all(dat), val, cmp); assert(itr != dat.end() && !cmp(val, *itr)); return itr - dat.begin(); } int lower_bound(const T& val) const { assert(sorted); auto itr = std::lower_bound(all(dat), val, cmp); return itr - dat.begin(); } int upper_bound(const T& val) const { assert(sorted); auto itr = std::upper_bound(all(dat), val, cmp); return itr - dat.begin(); } bool contains(const T& val) const { assert(sorted); return std::binary_search(all(dat), val, cmp); } std::vector<int> pressed(const std::vector<T>& vec) const { assert(sorted); std::vector<int> res(vec.size()); rep (i, vec.size()) res[i] = get(vec[i]); return res; } void press(std::vector<T>& vec) const { assert(sorted); for (auto&& i : vec) i = get(i); } int size() const { assert(sorted); return dat.size(); } }; #line 2 "library/math/PollardRho.hpp" #line 2 "library/random/Random.hpp" #line 4 "library/random/Random.hpp" template<class Engine> class Random { private: Engine rnd; public: using result_type = typename Engine::result_type; Random() : Random(std::random_device{}()) {} Random(result_type seed) : rnd(seed) {} result_type operator()() { return rnd(); } result_type min() const { return rnd.min(); } result_type max() const { return rnd.max(); } template<class IntType = ll> IntType uniform(IntType l, IntType r) { static_assert(std::is_integral<IntType>::value, "template argument must be an integral type"); assert(l <= r); return std::uniform_int_distribution<IntType>{l, r}(rnd); } template<class RealType = double> RealType uniform_real(RealType l, RealType r) { static_assert(std::is_floating_point<RealType>::value, "template argument must be an floating point type"); assert(l <= r); return std::uniform_real_distribution<RealType>{l, r}(rnd); } bool uniform_bool() { return uniform<int>(0, 1) == 1; } template<class T = ll> std::pair<T, T> uniform_pair(T l, T r) { assert(l < r); T a, b; do { a = uniform<T>(l, r); b = uniform<T>(l, r); } while (a == b); if (a > b) swap(a, b); return {a, b}; } template<class T = ll> std::vector<T> choice(int n, T l, T r) { assert(l <= r); assert(T(n) <= (r - l + 1)); std::set<T> res; while ((int)res.size() < n) res.insert(uniform<T>(l, r)); return {res.begin(), res.end()}; } template<class Iter> void shuffle(const Iter& first, const Iter& last) { std::shuffle(first, last, rnd); } template<class T> std::vector<T> permutation(T n) { std::vector<T> res(n); rep (i, n) res[i] = i; shuffle(all(res)); return res; } template<class T = ll> std::vector<T> choice_shuffle(int n, T l, T r, bool sorted = true) { assert(l <= r); assert(T(n) <= (r - l + 1)); std::vector<T> res(r - l + 1); rep (i, l, r + 1) res[i - l] = i; shuffle(all(res)); res.erase(res.begin() + n, res.end()); if (sorted) sort(all(res)); return res; } }; using Random32 = Random<std::mt19937>; Random32 rand32; using Random64 = Random<std::mt19937_64>; Random64 rand64; /** * @brief Random * @docs docs/random/Random.md */ #line 2 "library/math/MontgomeryModInt.hpp" #line 4 "library/math/MontgomeryModInt.hpp" template<class T> class MontgomeryReduction { static_assert(std::is_integral<T>::value, "T must be integral"); static_assert(std::is_unsigned<T>::value, "T must be unsigned"); private: using large_t = typename double_size_uint<T>::type; static constexpr int lg = std::numeric_limits<T>::digits; T mod; T r; T r2; // r^2 mod m T calc_minv() { T t = 0, res = 0; rep (i, lg) { if (~t & 1) { t += mod; res += static_cast<T>(1) << i; } t >>= 1; } return res; } T minv; public: MontgomeryReduction(T v) { set_mod(v); } static constexpr int get_lg() { return lg; } void set_mod(T v) { assert(v > 0); assert(v & 1); assert(v <= std::numeric_limits<T>::max() / 2); mod = v; r = (-static_cast<T>(mod)) % mod; r2 = (-static_cast<large_t>(mod)) % mod; minv = calc_minv(); } inline T get_mod() const { return mod; } inline T get_r() const { return r; } T reduce(large_t x) const { large_t tmp = (x + static_cast<large_t>(static_cast<T>(x) * minv) * mod) >> lg; return tmp >= mod ? tmp - mod : tmp; } T transform(large_t x) const { return reduce(x * r2); } }; template<class T, int id> class MontgomeryModInt { private: using large_t = typename double_size_uint<T>::type; using signed_t = typename std::make_signed<T>::type; T val; static MontgomeryReduction<T> mont; public: MontgomeryModInt() : val(0) {} template<class U, typename std::enable_if< std::is_integral<U>::value && std::is_unsigned<U>::value>::type* = nullptr> MontgomeryModInt(U x) : val(mont.transform( x < (static_cast<large_t>(mont.get_mod()) << mont.get_lg()) ? x : x % mont.get_mod())) {} template<class U, typename std::enable_if<std::is_integral<U>::value && std::is_signed<U>::value>::type* = nullptr> MontgomeryModInt(U x) : MontgomeryModInt(static_cast<typename std::make_unsigned<U>::type>( x < 0 ? -x : x)) { if (x < 0 && val) val = mont.get_mod() - val; } T get() const { return mont.reduce(val); } static T get_mod() { return mont.get_mod(); } static void set_mod(T v) { mont.set_mod(v); } MontgomeryModInt operator+() const { return *this; } MontgomeryModInt operator-() const { MontgomeryModInt res; if (val) res.val = mont.get_mod() - val; return res; } MontgomeryModInt& operator++() { val += mont.get_r(); if (val >= mont.get_mod()) val -= mont.get_mod(); return *this; } MontgomeryModInt& operator--() { if (val < mont.get_r()) val += mont.get_mod(); val -= mont.get_r(); return *this; } MontgomeryModInt operator++(int) { MontgomeryModInt res = *this; ++*this; return res; } MontgomeryModInt operator--(int) { MontgomeryModInt res = *this; --*this; return res; } MontgomeryModInt& operator+=(const MontgomeryModInt& rhs) { val += rhs.val; if (val >= mont.get_mod()) val -= mont.get_mod(); return *this; } MontgomeryModInt& operator-=(const MontgomeryModInt& rhs) { if (val < rhs.val) val += mont.get_mod(); val -= rhs.val; return *this; } MontgomeryModInt& operator*=(const MontgomeryModInt& rhs) { val = mont.reduce(static_cast<large_t>(val) * rhs.val); return *this; } MontgomeryModInt pow(ull n) const { MontgomeryModInt res = 1, x = *this; while (n) { if (n & 1) res *= x; x *= x; n >>= 1; } return res; } MontgomeryModInt inv() const { return pow(mont.get_mod() - 2); } MontgomeryModInt& operator/=(const MontgomeryModInt& rhs) { return *this *= rhs.inv(); } friend MontgomeryModInt operator+(const MontgomeryModInt& lhs, const MontgomeryModInt& rhs) { return MontgomeryModInt(lhs) += rhs; } friend MontgomeryModInt operator-(const MontgomeryModInt& lhs, const MontgomeryModInt& rhs) { return MontgomeryModInt(lhs) -= rhs; } friend MontgomeryModInt operator*(const MontgomeryModInt& lhs, const MontgomeryModInt& rhs) { return MontgomeryModInt(lhs) *= rhs; } friend MontgomeryModInt operator/(const MontgomeryModInt& lhs, const MontgomeryModInt& rhs) { return MontgomeryModInt(lhs) /= rhs; } friend bool operator==(const MontgomeryModInt& lhs, const MontgomeryModInt& rhs) { return lhs.val == rhs.val; } friend bool operator!=(const MontgomeryModInt& lhs, const MontgomeryModInt& rhs) { return lhs.val != rhs.val; } template<class Pr> void print(Pr& a) const { a.print(mont.reduce(val)); } template<class Pr> void debug(Pr& a) const { a.print(mont.reduce(val)); } template<class Sc> void scan(Sc& a) { ll v; a.scan(v); *this = v; } }; template<class T, int id> MontgomeryReduction<T> MontgomeryModInt<T, id>::mont = MontgomeryReduction<T>(998244353); using mmodint = MontgomeryModInt<unsigned int, -1>; /** * @brief MontgomeryModInt(モンゴメリ乗算) * @docs docs/math/MontgomeryModInt.md */ #line 2 "library/math/MillerRabin.hpp" #line 5 "library/math/MillerRabin.hpp" constexpr ull base_miller_rabin_int[3] = {2, 7, 61}; constexpr ull base_miller_rabin_ll[7] = {2, 325, 9375, 28178, 450775, 9780504, 1795265022}; template<class T> constexpr bool miller_rabin(ull n, const ull base[], int s) { if (T::get_mod() != n) T::set_mod(n); ull d = n - 1; while (~d & 1) d >>= 1; T e{1}, re{n - 1}; rep (i, s) { ull a = base[i]; if (a >= n) return true; ull t = d; T y = T(a).pow(t); while (t != n - 1 && y != e && y != re) { y *= y; t <<= 1; } if (y != re && !(t & 1)) return false; } return true; } constexpr bool is_prime_mr(ll n) { if (n == 2) return true; if (n < 2 || n % 2 == 0) return false; if (n < (1u << 31)) return miller_rabin<MontgomeryModInt<unsigned int, -2>>( n, base_miller_rabin_int, 3); return miller_rabin<MontgomeryModInt<ull, -2>>(n, base_miller_rabin_ll, 7); } #if __cpp_variable_templates >= 201304L && __cpp_constexpr >= 201304L template<ull n> constexpr bool is_prime_v = is_prime_mr(n); #endif /** * @brief MillerRabin(ミラーラビン素数判定) * @docs docs/math/MillerRabin.md */ #line 2 "library/string/RunLength.hpp" #line 4 "library/string/RunLength.hpp" template<class Cont, class Comp> std::vector<std::pair<typename Cont::value_type, int>> RunLength(const Cont& str, const Comp& cmp) { std::vector<std::pair<typename Cont::value_type, int>> res; if (str.size() == 0) return res; res.emplace_back(str[0], 1); rep (i, 1, str.size()) { if (cmp(res.back().first, str[i])) ++res.back().second; else res.emplace_back(str[i], 1); } return res; } template<class Cont> std::vector<std::pair<typename Cont::value_type, int>> RunLength(const Cont& str) { return RunLength(str, std::equal_to<typename Cont::value_type>()); } /** * @brief RunLength(ランレングス圧縮) * @docs docs/string/RunLength.md */ #line 8 "library/math/PollardRho.hpp" template<class T, class Rnd> ull pollard_rho(ull n, Rnd& rnd) { if (~n & 1) return 2; if (T::get_mod() != n) T::set_mod(n); T c, one = 1; auto f = [&](T x) -> T { return x * x + c; }; constexpr int M = 128; while (1) { c = rnd.uniform(1ull, n - 1); T x = rnd.uniform(2ull, n - 1), y = x; ull g = 1; while (g == 1) { T p = one, tx = x, ty = y; rep (M) { x = f(x); y = f(f(y)); p *= x - y; } g = gcd(p.get(), n); if (g == 1) continue; rep (M) { tx = f(tx); ty = f(f(ty)); g = gcd((tx - ty).get(), n); if (g != 1) { if (g != n) return g; break; } } } } return -1; } template<class T = MontgomeryModInt<ull, -3>, class Rnd = Random64> std::vector<ull> factorize(ull n, Rnd& rnd = rand64) { if (n == 1) return {}; std::vector<ull> res; std::vector<ull> st = {n}; while (!st.empty()) { ull t = st.back(); st.pop_back(); if (t == 1) continue; if (is_prime_mr(t)) { res.push_back(t); continue; } ull f = pollard_rho<T>(t, rnd); st.push_back(f); st.push_back(t / f); } std::sort(all(res)); return res; } template<class T = MontgomeryModInt<ull, -3>, class Rnd = Random64> std::vector<std::pair<ull, int>> expfactorize(ull n, Rnd& rnd = rand64) { auto f = factorize<T, Rnd>(n, rnd); return RunLength(f); } std::vector<ll> divisors_pr(ll n) { std::vector<ll> res; auto r = expfactorize(n); int m = r.size(); rec_lambda([&](auto&& self, int k, ll d) -> void { if (k == m) { res.push_back(d); return; } ll t = d; rep (r[k].second) { self(k + 1, d); d *= r[k].first; } self(k + 1, d); d = t; })(0, 1); std::sort(all(res)); return res; } /** * @brief PollardRho(素因数分解) * @docs docs/math/PollardRho.md */ #line 3 "main.cpp" using namespace std; int main() { ll N; scan >> N; vector<map<ll, int>> memo(61); ll ans = 0; rep (len, 1, 60) { ans += rec_lambda([&](auto&& self, int i, ll n) -> ll { if (n == 1) return 1; if (i == len) return 0; if (memo[len - i].count(n)) return memo[len - i][n]; auto ef = expfactorize(n); vector<ll> a{1}; for (auto [p, e] : ef) { e /= len - i; vector<ll> b; for (auto i : a) { ll t = 1; rep (e + 1) { b.push_back(i * t); t *= p; } } a = move(b); } ll ans = 0; for (auto d : a) { if (i == 0 && d == 1) continue; ans += self(i + 1, n / my_pow(d, len - i)); } return memo[len - i][n] = ans; })(0, N); } prints(ans); }