結果
問題 | No.577 Prime Powerful Numbers |
ユーザー | firiexp |
提出日時 | 2020-03-06 01:09:09 |
言語 | C++14 (gcc 12.3.0 + boost 1.83.0) |
結果 |
AC
|
実行時間 | 1,365 ms / 2,000 ms |
コード長 | 6,113 bytes |
コンパイル時間 | 1,275 ms |
コンパイル使用メモリ | 109,100 KB |
実行使用メモリ | 5,248 KB |
最終ジャッジ日時 | 2024-11-15 19:16:01 |
合計ジャッジ時間 | 5,769 ms |
ジャッジサーバーID (参考情報) |
judge5 / judge2 |
(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 | 22 ms
5,248 KB |
testcase_01 | AC | 78 ms
5,248 KB |
testcase_02 | AC | 12 ms
5,248 KB |
testcase_03 | AC | 245 ms
5,248 KB |
testcase_04 | AC | 82 ms
5,248 KB |
testcase_05 | AC | 1,354 ms
5,248 KB |
testcase_06 | AC | 61 ms
5,248 KB |
testcase_07 | AC | 1,365 ms
5,248 KB |
testcase_08 | AC | 224 ms
5,248 KB |
testcase_09 | AC | 216 ms
5,248 KB |
testcase_10 | AC | 4 ms
5,248 KB |
ソースコード
#include <limits> #include <iostream> #include <algorithm> #include <iomanip> #include <map> #include <set> #include <queue> #include <stack> #include <numeric> #include <bitset> #include <cmath> static const int MOD = 1000000007; using ll = long long; using u32 = unsigned; using u64 = unsigned long long; using namespace std; template<class T> constexpr T INF = ::numeric_limits<T>::max()/32*15+208; #include <random> using u128 = __uint128_t; struct mod64 { u64 n; static u64 mod, inv, r2; mod64() : n(0) {} mod64(u64 x) : n(init(x)) {} static u64 init(u64 w) { return reduce(u128(w) * r2); } static void set_mod(u64 m) { mod = inv = m; for (int i = 0; i < 5; ++i) inv *= 2 - inv * m; r2 = -u128(m) % m; } static u64 reduce(u128 x) { u64 y = u64(x >> 64) - u64((u128(u64(x) * inv) * mod) >> 64); return ll(y) < 0 ? y + mod : y; }; mod64& operator+=(mod64 x) { n += x.n - mod; if(ll(n) < 0) n += mod; return *this; } mod64 operator+(mod64 x) const { return mod64(*this) += x; } mod64& operator*=(mod64 x) { n = reduce(u128(n) * x.n); return *this; } mod64 operator*(mod64 x) const { return mod64(*this) *= x; } u64 val() const { return reduce(n); } }; u64 mod64::mod, mod64::inv, mod64::r2; bool suspect(u64 a, u64 s, u64 d, u64 n){ mod64::set_mod(n); mod64 x(1), xx(a), one(x), minusone(n-1); while(d > 0){ if(d&1) x = x * xx; xx = xx * xx; d >>= 1; } if (x.n == one.n) return true; for (int r = 0; r < s; ++r) { if(x.n == minusone.n) return true; x = x * x; } return false; } template<class T> bool miller_rabin(T n){ if (n <= 1 || (n > 2 && n % 2 == 0)) return false; u64 d = n - 1, s = 0; while (!(d&1)) {++s; d >>= 1;} static const u64 v[] = {2, 325, 9375, 28178, 450775, 9780504, 1795265022}; static const u64 v_small[] = {2, 7, 61}; if(n <= 4759123141LL){ for (auto &&p : v_small) { if(p >= n) break; if(!suspect(p, s, d, n)) return false; } }else { for (auto &&p : v) { if(p >= n) break; if(!suspect(p, s, d, n)) return false; } } return true; } template<typename T> struct ExactDiv { T t, i, val; ExactDiv() {} ExactDiv(T n) : t(T(-1) / n), i(mul_inv(n)) , val(n) {}; T mul_inv(T n) { T x = n; for (int i = 0; i < 5; ++i) x *= 2 - n * x; return x; } bool divide(T n) const { if(val == 2) return !(n & 1); return n * this->i <= this->t; } }; vector<ExactDiv<u64>> get_prime(int n){ if(n <= 1) return vector<ExactDiv<u64>>(); vector<bool> is_prime(n+1, true); vector<ExactDiv<u64>> prime; is_prime[0] = is_prime[1] = false; for (int i = 2; i <= n; ++i) { if(is_prime[i]) prime.emplace_back(i); for (auto &&j : prime){ if(i*j.val > n) break; is_prime[i*j.val] = false; if(j.divide(i)) break; } } return prime; } const auto primes = get_prime(32000); random_device rng; template<class T> T pollard_rho2(T n) { uniform_int_distribution<T> ra(1, n-1); mod64::set_mod(n); while(true){ u64 c_ = ra(rng), g = 1, r = 1, m = 500; while(c_ == n-2) c_ = ra(rng); mod64 y(ra(rng)), xx(0), c(c_), ys(0), q(1); while(g == 1){ xx.n = y.n; for (int i = 1; i <= r; ++i) { y *= y; y += c; } T k = 0; g = 1; while(k < r && g == 1){ for (int i = 1; i <= (m > (r-k) ? (r-k) : m); ++i) { ys.n = y.n; y *= y; y += c; u64 xxx = xx.val(), yyy = y.val(); q *= mod64(xxx > yyy ? xxx - yyy : yyy - xxx); } g = __gcd<ll>(q.val(), n); k += m; } r *= 2; } if(g == n) g = 1; while (g == 1){ ys *= ys; ys += c; u64 xxx = xx.val(), yyy = ys.val(); g = __gcd<ll>(xxx > yyy ? xxx - yyy : yyy - xxx, n); } if (g != n && miller_rabin(g)) return g; } } template<class T> vector<T> prime_factor(T n, int d = 0){ vector<T> a, res; if(!d) for (auto &&i : primes) { while (i.divide(n)){ res.emplace_back(i.val); n /= i.val; } } while(n != 1){ if(miller_rabin(n)){ a.emplace_back(n); break; } T x = pollard_rho2(n); n /= x; a.emplace_back(x); } for (auto &&i : a) { if (miller_rabin(i)) { res.emplace_back(i); } else { vector<T> b = prime_factor(i, d + 1); for (auto &&j : b) res.emplace_back(j); } } sort(res.begin(),res.end()); return res; } void solve(){ } int main() { int q; cin >> q; auto f = [&](ll x){ if(x == 2) return false; if(x%2 == 0) return true; for (ll p = 2; p+2 <= x; p *= 2) { if(miller_rabin(x-p)) return true; ll ok = 0, ng = 1000000001; // p^2 while(ng-ok > 1){ ll mid = (ok+ng)/2; if(mid*mid <= x-p) ok = mid; else ng = mid; } if(ok*ok == x-p && miller_rabin(ok)) return true; ok = 0, ng = 1000001; while(ng-ok > 1){ ll mid = (ok+ng)/2; if(mid*mid*mid <= x-p) ok = mid; else ng = mid; } if(ok*ok*ok == x-p && miller_rabin(ok)) return true; for (auto &&i : primes) { u64 pp = i.val*i.val*i.val; for (int j = 4; j*log10(i.val) < 18.5; ++j) { pp *= i.val; if(pp == x-p) return true; } } } return false; }; while(q--){ ll x; cin >> x; puts(f(x) ? "Yes" : "No"); } return 0; }