Submission #927812

# Submission time Handle Problem Language Result Execution time Memory
927812 2024-02-15T11:01:48 Z cadmiumsky Segments (IZhO18_segments) C++17
7 / 100
1492 ms 18300 KB
#pragma GCC optimize("O3,unroll-loops")
#include <bits/stdc++.h>
#define all(x) (x).begin(),(x).end()
using namespace std;

using ll = long long;
using ld = long double;

//#define int ll
#define sz(x) ((int)(x).size())

using pii = pair<int,int>;
using tii = tuple<int,int,int>;

const int nmax = 2e5 + 5;

pii segments[nmax];

struct Oper {
  char type;
  int a, b;
  int k;
};

namespace Precalc {
  multiset<pii> actives_sortfwr;
  
  auto cmp = [](const pii& a, const pii& b) -> bool { return a.second < b.second || (a.second == b.second && a.first < b.first); };
  multiset<pii, decltype(cmp)> actives_sortbckw(cmp);
  
  void rebatch(vector<Oper> opqueue) {
    for(auto [t, a, b, k] : opqueue) {
      if(t == 1)
        actives_sortfwr.insert(pii{a, b}),
        actives_sortbckw.insert(pii{a, b});
      else if(t == 2)
        actives_sortfwr.erase(actives_sortfwr.find(segments[a])),
        actives_sortbckw.erase(actives_sortbckw.find(segments[a]));
    }
    
    return;
  }
  struct Item {
    int link, sum, own;
  };
  
  set<pii> fwrpoz;
  set<pii> bckwpoz;
  int freq[nmax];
  Item fwr[nmax], bckw[nmax];
  pii atrf[nmax], atrb[nmax];
  #warning caching, pune-le impreuna?
  #warning si gen, deseori prinzi chestia din mijlocul prefixului de k-uri interesante, poti ""injumatati"" timpul daca faci chestii care tin cont de asta
  vector<int> currKs;
  
  int n;
  
  void contextualize(vector<int> Ks) {    
    fwrpoz.clear(), bckwpoz.clear();
    
    { // forward links
      pii lst = {-1, -1};
      
      n = 0;
      for(auto [a, b] : actives_sortfwr) {
        int k = b - a + 1;
        if(k < Ks[0]) continue;
        if(pii{a, b} != lst) { n++; fwr[n].sum = fwr[n].own = 0; }
        atrf[n] = lst = pii{a, b};
        fwr[n].own++;
      }
      
      vector<int> counter_app(sz(Ks));
      
      for(int i = n, atrcoef = sz(Ks) - 1; i > 0; atrcoef = (atrcoef == 0? sz(Ks) - 1 : atrcoef - 1), i--) {
        auto [a, b] = atrf[i];
        
        int poz = distance(begin(Ks), prev(upper_bound(all(Ks), b - a + 1)));
        counter_app[poz] += fwr[i].own;
        
        fwr[i].link = atrcoef;
        fwr[i].sum = counter_app[atrcoef];
        
        fwrpoz.emplace(pii{a, i});
      }
    }
    // ---- 
    
    { // backward links
      pii lst = {-1, -1};
      
      n = 0;
      for(auto [a, b] : actives_sortbckw) {
        int k = b - a + 1;
        if(k < Ks[0]) continue;
        if(pii{a, b} != lst) { n++; bckw[n].sum = bckw[n].own = 0; }
        atrb[n] = lst = pii{a, b};
        bckw[n].own++;
      }
      
      vector<int> counter_app(sz(Ks));
      
      for(int i = 1, atrcoef = sz(Ks) - 1; i <= n; atrcoef = (atrcoef == 0? sz(Ks) - 1 : atrcoef - 1), i++) {
        auto [a, b] = atrb[i];
        
        int poz = distance(begin(Ks), prev(upper_bound(all(Ks), b - a + 1)));
        counter_app[poz] += bckw[i].own;
        
        bckw[i].link = atrcoef;
        bckw[i].sum = counter_app[atrcoef];
        
        bckwpoz.emplace(pii{b, i});
      }
    }
      
    for(auto [a, b] : actives_sortfwr) {
      int k = b - a + 1;
      if(k < Ks[0]) continue;
      freq[distance(begin(Ks), prev(upper_bound(all(Ks), k)))]++;
    }
    
    for(int i = sz(Ks) - 2; i >= 0; i--)
      freq[i] += freq[i + 1];
    currKs = move(Ks);
    
  }
  
  int query_K(int k) { 
    k = distance(begin(currKs), lower_bound(all(currKs), k));
    return freq[k];
  }
  int query_L(int L, int k) {
    k = distance(begin(currKs), lower_bound(all(currKs), k));
    auto it = bckwpoz.upper_bound(pii{L + 1, - 1});
    if(it == bckwpoz.begin()) return 0;
    
    int poz = prev(it) -> second;
    int rez = 0;
    while(k < sz(currKs) && poz > 0) {
      if(bckw[poz].link == k) {
        rez += bckw[poz].sum;
        k++;
      }
      
      if(k < sz(currKs) && currKs[k] <= atrb[poz].second - atrb[poz].first + 1) rez += bckw[poz].own;
      poz--;
    }
    return rez;
  }
  
  int query_R(int R, int k) {
    k = distance(begin(currKs), lower_bound(all(currKs), k));
    auto it = fwrpoz.upper_bound(pii{R, - 1});
    if(it == fwrpoz.end()) return 0;
    
    int poz = it -> second;
    int rez = 0;
    while(k < sz(currKs) && poz <= n) {
      if(fwr[poz].link == k) {
        rez += fwr[poz].sum;
        k++;
      }
      
      if(k < sz(currKs) && currKs[k] <= atrf[poz].second - atrf[poz].first + 1) rez += fwr[poz].own;
      poz++;
    }
    return rez;
  }
} 

//#error in loc sa ai pozitii care corespund unor segmente bune, doar pune suma partiala a alora congruenti cu (-pozitia) mod sz(Ks)

int segmpoz = 1;

int unitValue(int l, int r, int a, int b, int k) {
  return min(r, b) - max(l, a) + 1 >= k;
}

signed main() {
  cin.tie(0) -> sync_with_stdio(0);
  int n, t_offline;
  cin >> n >> t_offline;
  
  const int idealBuck = 625;
  const ll worstInner = (ll)idealBuck * n + 5;
  
  
  vector<Oper> operation_queue;
  operation_queue.reserve(n);
  vector<int> results;
  results.reserve(n);

  int lastans = 0;
  
  auto flush_queue = [&]() {
    vector<int> Ks;
    Ks.reserve(sz(operation_queue));
    for(auto [t, a, b, k] : operation_queue)
      if(t == 3) Ks.emplace_back(k);
    
    if(sz(Ks) == 0) return;
    
    sort(all(Ks));
    Ks.erase(unique(all(Ks)), Ks.end());
    
    Precalc::contextualize(Ks);
    
    for(int i = 0; i < sz(operation_queue); i++) {
      auto &[ti, ai, bi, ki] = operation_queue[i];
      if(ti == 1 || ti == 3) {
        ai = ai ^ (t_offline * lastans);
        bi = bi ^ (t_offline * lastans);
        if(ai > bi) swap(ai, bi);
      }
      
      if(ti == 1)
        segments[segmpoz++] = pii{ai, bi};
      
      if(ti == 3) {
        if(bi - ai + 1 < ki) {
          lastans = 0;
        }
        else {
          int rez = Precalc::query_K(ki) - Precalc::query_R(bi - ki + 1 + 1, ki) - Precalc::query_L(ai + ki - 1 - 1, ki);
          for(int j = i - 1; j >= 0; j--) {
            auto &[tj, aj, bj, kj] = operation_queue[j];
            
            if(tj == 3) continue;
            else if(tj == 2)
              rez -= unitValue(ai, bi, segments[aj].first, segments[aj].second, ki);
            else
              rez += unitValue(ai, bi, aj, bj, ki);
          }
          lastans = rez;
        }
        
        results.emplace_back(lastans);
      }
    }
    
    Precalc::rebatch(operation_queue);
    operation_queue.clear();
  };
  int cnt[2] = {1, 0};
  
  for(int t, a, b, k, tc = 0; tc < n; tc++) {
    cin >> t;
    if(t == 1) {
       cin >> a >> b;
       operation_queue.emplace_back(Oper{t, a, b, -1});
       cnt[0]++;
    }
    if(t == 2) {
       cin >> a;
       operation_queue.emplace_back(Oper{t, a, -1, -1});
       cnt[0]++;
    }
    if(t == 3) {
       cin >> a >> b >> k;
       operation_queue.emplace_back(Oper{t, a, b, k});
       cnt[1]++;
    }
    
    if(tc == n - 1 || (ll)cnt[0] * cnt[1] >= worstInner) {
      flush_queue();
      
      cnt[0] = sz(Precalc::actives_sortfwr);
      cnt[1] = 0;
    }
  }
  
  for(auto x : results) cout << x << '\n';
  return 0;
}

/**
  Anul asta nu se da centroid
  -- Rugaciunile mele
*/

Compilation message

segments.cpp:52:4: warning: #warning caching, pune-le impreuna? [-Wcpp]
   52 |   #warning caching, pune-le impreuna?
      |    ^~~~~~~
segments.cpp:53:4: warning: #warning si gen, deseori prinzi chestia din mijlocul prefixului de k-uri interesante, poti ""injumatati"" timpul daca faci chestii care tin cont de asta [-Wcpp]
   53 |   #warning si gen, deseori prinzi chestia din mijlocul prefixului de k-uri interesante, poti ""injumatati"" timpul daca faci chestii care tin cont de asta
      |    ^~~~~~~
segments.cpp: In function 'int main()':
segments.cpp:250:42: warning: narrowing conversion of 't' from 'int' to 'char' [-Wnarrowing]
  250 |        operation_queue.emplace_back(Oper{t, a, b, -1});
      |                                          ^
segments.cpp:255:42: warning: narrowing conversion of 't' from 'int' to 'char' [-Wnarrowing]
  255 |        operation_queue.emplace_back(Oper{t, a, -1, -1});
      |                                          ^
segments.cpp:260:42: warning: narrowing conversion of 't' from 'int' to 'char' [-Wnarrowing]
  260 |        operation_queue.emplace_back(Oper{t, a, b, k});
      |                                          ^
# Verdict Execution time Memory Grader output
1 Correct 0 ms 344 KB Output is correct
2 Correct 0 ms 348 KB Output is correct
3 Correct 7 ms 6748 KB Output is correct
4 Correct 7 ms 6748 KB Output is correct
5 Correct 10 ms 7516 KB Output is correct
6 Correct 11 ms 7260 KB Output is correct
7 Correct 8 ms 6724 KB Output is correct
8 Correct 7 ms 860 KB Output is correct
9 Correct 8 ms 7256 KB Output is correct
10 Correct 4 ms 860 KB Output is correct
11 Correct 15 ms 7256 KB Output is correct
12 Correct 14 ms 7260 KB Output is correct
13 Correct 4 ms 860 KB Output is correct
14 Correct 8 ms 7000 KB Output is correct
15 Correct 7 ms 6744 KB Output is correct
16 Correct 7 ms 6724 KB Output is correct
17 Correct 8 ms 7004 KB Output is correct
18 Correct 5 ms 860 KB Output is correct
19 Correct 8 ms 7004 KB Output is correct
20 Correct 8 ms 7004 KB Output is correct
# Verdict Execution time Memory Grader output
1 Incorrect 1492 ms 18300 KB Output isn't correct
2 Halted 0 ms 0 KB -
# Verdict Execution time Memory Grader output
1 Incorrect 490 ms 7964 KB Output isn't correct
2 Halted 0 ms 0 KB -
# Verdict Execution time Memory Grader output
1 Incorrect 497 ms 7756 KB Output isn't correct
2 Halted 0 ms 0 KB -
# Verdict Execution time Memory Grader output
1 Correct 0 ms 344 KB Output is correct
2 Correct 0 ms 348 KB Output is correct
3 Correct 7 ms 6748 KB Output is correct
4 Correct 7 ms 6748 KB Output is correct
5 Correct 10 ms 7516 KB Output is correct
6 Correct 11 ms 7260 KB Output is correct
7 Correct 8 ms 6724 KB Output is correct
8 Correct 7 ms 860 KB Output is correct
9 Correct 8 ms 7256 KB Output is correct
10 Correct 4 ms 860 KB Output is correct
11 Correct 15 ms 7256 KB Output is correct
12 Correct 14 ms 7260 KB Output is correct
13 Correct 4 ms 860 KB Output is correct
14 Correct 8 ms 7000 KB Output is correct
15 Correct 7 ms 6744 KB Output is correct
16 Correct 7 ms 6724 KB Output is correct
17 Correct 8 ms 7004 KB Output is correct
18 Correct 5 ms 860 KB Output is correct
19 Correct 8 ms 7004 KB Output is correct
20 Correct 8 ms 7004 KB Output is correct
21 Incorrect 1492 ms 18300 KB Output isn't correct
22 Halted 0 ms 0 KB -
# Verdict Execution time Memory Grader output
1 Correct 0 ms 344 KB Output is correct
2 Correct 0 ms 348 KB Output is correct
3 Correct 7 ms 6748 KB Output is correct
4 Correct 7 ms 6748 KB Output is correct
5 Correct 10 ms 7516 KB Output is correct
6 Correct 11 ms 7260 KB Output is correct
7 Correct 8 ms 6724 KB Output is correct
8 Correct 7 ms 860 KB Output is correct
9 Correct 8 ms 7256 KB Output is correct
10 Correct 4 ms 860 KB Output is correct
11 Correct 15 ms 7256 KB Output is correct
12 Correct 14 ms 7260 KB Output is correct
13 Correct 4 ms 860 KB Output is correct
14 Correct 8 ms 7000 KB Output is correct
15 Correct 7 ms 6744 KB Output is correct
16 Correct 7 ms 6724 KB Output is correct
17 Correct 8 ms 7004 KB Output is correct
18 Correct 5 ms 860 KB Output is correct
19 Correct 8 ms 7004 KB Output is correct
20 Correct 8 ms 7004 KB Output is correct
21 Incorrect 1492 ms 18300 KB Output isn't correct
22 Halted 0 ms 0 KB -