#pragma GCC optimize("Ofast")
#include <bits/stdc++.h>
using namespace std;
int LIM = 2000;
int in[200005];
int out[200005];
int region[200005];
int parent[200005];
int cnt[200005];
vector <int> graf[200005];
int gore[25005][505];
int dole[25005][505];
int sada[25005];
int rev[25005];
int compress[25005];
int ccomp;
int r;
int tajm;
vector <pair <int, int>> svi[25005];
vector <int> ins[25005];
void dfs(int v){
in[v] = ++tajm;
sada[region[v]]++;
if(compress[region[v]]){
for(int i=1; i<=r; i++){
dole[i][compress[region[v]]] += sada[i];
}
dole[region[v]][compress[region[v]]]--;
gore[region[v]][compress[region[v]]]--;
}
for(int i=1; i<=ccomp; i++){
gore[region[v]][i] += sada[rev[i]];
}
for(auto c : graf[v]){
dfs(c);
}
sada[region[v]]--;
out[v] = tajm;
svi[region[v]].push_back({in[v], 1});
svi[region[v]].push_back({out[v]+1, -1});
ins[region[v]].push_back(in[v]);
}
int main(){
ios_base::sync_with_stdio(false), cin.tie(0);
int n, q;
cin >> n >> r >> q;
cin >> region[1];
cnt[region[1]]++;
for(int i=2; i<=n; i++){
cin >> parent[i] >> region[i];
cnt[region[i]]++;
graf[parent[i]].push_back(i);
}
for(int i=1; i<=r; i++){
if(cnt[i] <= LIM) continue;
compress[i] = ++ccomp;
rev[ccomp] = i;
}
dfs(1);
for(int i=1; i<=r; i++) sort(svi[i].begin(), svi[i].end());
for(int i=1; i<=r; i++) sort(ins[i].begin(), ins[i].end());
while(q--){
int a, b;
cin >> a >> b;
if(cnt[a] > LIM){
cout << gore[b][compress[a]] << endl;
continue;
}
if(cnt[b] > LIM){
cout << dole[a][compress[b]] << endl;
continue;
}
int k = svi[a].size();
int j = -1;
int tren = 0;
int res = 0;
for(auto c : ins[b]){
while(j < k-1 && c >= svi[a][j+1].first){
j++;
tren += svi[a][j].second;
}
res += tren;
}
cout << res << endl;
}
return 0;
}
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
5 ms |
6272 KB |
Output is correct |
2 |
Correct |
5 ms |
6272 KB |
Output is correct |
3 |
Correct |
9 ms |
6176 KB |
Output is correct |
4 |
Correct |
8 ms |
6272 KB |
Output is correct |
5 |
Correct |
11 ms |
6272 KB |
Output is correct |
6 |
Correct |
21 ms |
6400 KB |
Output is correct |
7 |
Correct |
37 ms |
6400 KB |
Output is correct |
8 |
Correct |
37 ms |
6400 KB |
Output is correct |
9 |
Correct |
76 ms |
6912 KB |
Output is correct |
10 |
Correct |
68 ms |
7040 KB |
Output is correct |
11 |
Correct |
113 ms |
7468 KB |
Output is correct |
12 |
Correct |
126 ms |
8064 KB |
Output is correct |
13 |
Correct |
131 ms |
7936 KB |
Output is correct |
14 |
Correct |
176 ms |
8704 KB |
Output is correct |
15 |
Correct |
205 ms |
11392 KB |
Output is correct |
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
867 ms |
13168 KB |
Output is correct |
2 |
Correct |
1006 ms |
12272 KB |
Output is correct |
3 |
Correct |
1642 ms |
15504 KB |
Output is correct |
4 |
Correct |
237 ms |
8576 KB |
Output is correct |
5 |
Correct |
326 ms |
10240 KB |
Output is correct |
6 |
Correct |
711 ms |
10232 KB |
Output is correct |
7 |
Correct |
933 ms |
11640 KB |
Output is correct |
8 |
Correct |
930 ms |
17000 KB |
Output is correct |
9 |
Correct |
1669 ms |
19364 KB |
Output is correct |
10 |
Correct |
2136 ms |
23544 KB |
Output is correct |
11 |
Correct |
2656 ms |
19908 KB |
Output is correct |
12 |
Correct |
4067 ms |
83516 KB |
Output is correct |
13 |
Correct |
4670 ms |
83728 KB |
Output is correct |
14 |
Correct |
6723 ms |
99096 KB |
Output is correct |
15 |
Execution timed out |
8026 ms |
124072 KB |
Time limit exceeded |
16 |
Correct |
6820 ms |
128892 KB |
Output is correct |
17 |
Correct |
6980 ms |
107868 KB |
Output is correct |