Submission #799975

# Submission time Handle Problem Language Result Execution time Memory
799975 2023-08-01T09:03:05 Z 이동현(#10084) Harvest (JOI20_harvest) C++17
0 / 100
5000 ms 482536 KB
#include <bits/stdc++.h>
#pragma GCC optimize("O3")
#pragma GCC optimize("Ofast")
#pragma GCC optimize("unroll-loops")
#define int long long
using namespace std;

int INF = (int)1e18;
struct Node{
    int l, r, cnt;
    Node(){
        l = r = cnt = 0;
    }
};
int tn;
Node tr[(int)1e7];

void push(int x, int s, int e, int pos){
    if(s == e){
        ++tr[x].cnt;
        return;
    }

    if(pos <= (s + e) / 2){
        if(!tr[x].l) tr[x].l = ++tn;
        push(tr[x].l, s, (s + e) / 2, pos);
    }
    else{
        if(!tr[x].r) tr[x].r = ++tn;
        push(tr[x].r, (s + e) / 2 + 1, e, pos);
    }
    tr[x].cnt = tr[tr[x].l].cnt + tr[tr[x].r].cnt;
}

int get(int x, int s, int e, int fe){
    if(s > fe) return 0;
    if(e <= fe){
        return tr[x].cnt;
    }

    int rv = 0;
    if(tr[x].l) rv += get(tr[x].l, s, (s + e) / 2, fe);
    if(tr[x].r && (s + e) / 2 + 1 <= fe) rv += get(tr[x].r, (s + e) / 2 + 1, e, fe);
    return rv;
}

signed main(){
    ios_base::sync_with_stdio(false);
    cin.tie(0);

    int n, m, l, c;
    cin >> n >> m >> l >> c;
    vector<int> a(n), b(m);
    for(int i = 0; i < n; ++i){
        cin >> a[i];
    }
    for(int i = 0; i < m; ++i){
        cin >> b[i];
    }

    auto god = [&](int x, int y){
        return c + (y - c % l - x + l * 2) % l;
    };

    vector<vector<int>> mine(n);
    vector<int> madd(n);
    int p = 0;
    for(int i = 0; i < n; ++i){
        while(p < m && a[i] > b[p]) ++p;

        int r = p - 1;
        while(r + 1 < m && (i == n - 1 || b[r + 1] < a[i + 1])){
            ++r;
            mine[i].push_back(b[r] - a[i]);
        }

        p = r;
    }

    for(int i = 0; i < m && b[i] < a[0]; ++i){
        mine[n - 1].push_back(b[i] - a[n - 1] + l);
    }

    vector<int> got(n);
    vector<vector<int>> way(n);
    for(int i = 0; i < n; ++i){
        int pos = (a[i] - c % l + l) % l;

        int p = lower_bound(a.begin(), a.end(), pos + 1) - a.begin();
        --p;
        if(p < 0) p = n - 1;

        got[i] = p;
        way[p].push_back(i);
    }

    int q;
    cin >> q;
    vector<int> ans(q);
    vector<vector<array<int, 2>>> que(n);
    for(int i = 0; i < q; ++i){
        int x, y;
        cin >> x >> y;
        --x;
        que[x].push_back({y, i});
    }

    vector<int> cyclen(n, -1);
    vector<int> chk(n), cdist(n);
    int cn = 0;

    auto checkcycle = [&](auto&&self, int x)->void{
        if(chk[x]) return;
        chk[x] = cn;

        if(chk[got[x]] == chk[x]){
            int val = cdist[x] - cdist[got[x]] + god(a[got[x]], a[x]);
            while(cyclen[x] == -1){
                cyclen[x] = val;
                x = got[x];
            }
            return;
        }

        cdist[got[x]] = cdist[x] + god(a[got[x]], a[x]);
        self(self, got[x]);
    };

    for(int i = 0; i < n; ++i){
        ++cn;
        checkcycle(checkcycle, i);
    }

    vector<int> ans2(q);
    for(int i = 0; i < n; ++i){
        vector<int> chk(n);

        auto sol = [&](auto&&self, int x, int dist)->void{
            if(chk[x]) return;
            chk[x] = 1;

            for(auto&[t, anspos]:que[i]){
                for(auto&j:mine[x]){
                    if(j + dist > t) continue;

                    ans2[anspos] += 1 + (cyclen[i] != -1 ? (t - (j + dist)) / cyclen[i] : 0);
                }
            }

            for(auto&nxt:way[x]){
                self(self, nxt, dist + god(a[x], a[nxt]));
            }
        };

        sol(sol, i, 0);
    }

    vector<int> chk1(n);
    vector<int> rt(n);
    iota(rt.begin(), rt.end(), 1);
    tn = n;
    auto sol1 = [&](auto&&self, int x)->void{
        // assert(cyclen[x] == -1);
        if(chk1[x]) return;
        chk1[x] = 1;

        if(cyclen[x] == -1){
            for(auto&i:mine[x]){
                push(rt[x], 1, INF, i);
            }
        }

        for(auto&nxt:way[x]){
            if(cyclen[nxt] != -1) continue;
            self(self, nxt);

            madd[nxt] += god(a[x], a[nxt]);

            assert(madd[x] >= 0 && madd[nxt] >= 0);

            // if((int)mine[x].size() < (int)mine[nxt].size()){
            //     swap(mine[x], mine[nxt]);
            //     swap(madd[x], madd[nxt]);
            //     swap(rt[x], rt[nxt]);
            // }

            while((int)mine[nxt].size()){
                int val = mine[nxt].back() + madd[nxt] - madd[x];
                mine[x].push_back(val);
                if(cyclen[x] == -1 && val + madd[x] <= INF){
                    push(rt[x], 1, INF, val);
                }
                mine[nxt].pop_back();
            }
        }

        if(cyclen[x] == -1){
            for(auto&[t, anspos]:que[x]){
                ans[anspos] += get(rt[x], 1, INF, t + madd[x]);
                // for(auto&j:mine[x]){
                //     if(j + madd[x] > t) continue;

                //     ans[anspos] += 1;
                // }
            }
        }
    };

    for(int i = 0; i < n; ++i){
        if(cyclen[i] != -1) continue;

        sol1(sol1, i);
    }

    for(int i = 0; i < n; ++i){
        if(cyclen[i] == -1) continue;

        sol1(sol1, i);

        sort(mine[i].begin(), mine[i].end());
        for(auto&j:mine[i]){
            j += madd[i];
        }
        while((int)mine[i].size() && mine[i].back() > INF){
            mine[i].pop_back();
        }
    }

    vector<int> didcycle(n);
    for(int i = 0; i < n; ++i){
        if(cyclen[i] == -1) continue;

        vector<int> cycle;
        int now = i;
        while(true){
            cycle.push_back(now);

            int nxt = -1;
            for(auto&j:way[now]){
                if(cyclen[j] != -1){
                    nxt = j;
                    break;
                }
            }
            assert(nxt != -1);

            now = nxt;

            if(now == i) break;
        }

        int dsum = 0;
        int nrt = ++tn;
        for(int j = 0; j < (int)cycle.size(); ++j){
            for(auto&x:mine[cycle[j]]){
                for(auto&[t, anspos]:que[i]){
                    int val = x + dsum;

                    if(val <= t) ans[anspos] += 1 + (t - val) / cyclen[i];
                }
                // push(nrt, 1, INF, x + dsum);
            }
            if(j + 1 < (int)cycle.size()){
                dsum += god(a[cycle[j]], a[cycle[j + 1]]);
            }
        }

        // for(int j = 0; j < (int)cycle.size(); ++j){
        //     for(auto&[t, anspos]:que[cycle[j]]){
        //         ans[anspos] += get(nrt, 1, INF, t);
        //     }
        // }
    }

    for(int i = 0; i < q; ++i){
        cout << ans[i] << '\n';
    }

    for(int i = 0; i < n; ++i){
        if(cyclen[i] != -1) continue;

        for(auto&[t, anspos]:que[i]){
            assert(ans[anspos] == ans2[anspos]);
        }
    }
    
    return 0;
}

Compilation message

harvest.cpp: In function 'int main()':
harvest.cpp:253:13: warning: unused variable 'nrt' [-Wunused-variable]
  253 |         int nrt = ++tn;
      |             ^~~
# Verdict Execution time Memory Grader output
1 Correct 92 ms 235244 KB Output is correct
2 Correct 89 ms 235792 KB Output is correct
3 Correct 246 ms 235820 KB Output is correct
4 Correct 177 ms 237792 KB Output is correct
5 Runtime error 612 ms 482536 KB Execution killed with signal 11
6 Halted 0 ms 0 KB -
# Verdict Execution time Memory Grader output
1 Execution timed out 5093 ms 242968 KB Time limit exceeded
2 Halted 0 ms 0 KB -
# Verdict Execution time Memory Grader output
1 Correct 92 ms 235244 KB Output is correct
2 Correct 89 ms 235792 KB Output is correct
3 Correct 246 ms 235820 KB Output is correct
4 Correct 177 ms 237792 KB Output is correct
5 Runtime error 612 ms 482536 KB Execution killed with signal 11
6 Halted 0 ms 0 KB -