#include <bits/stdc++.h>
using namespace std;
#pragma GCC optimize ("Ofast")
const int MAXN = 4e5 + 25;
int in[MAXN], out[MAXN], depth[MAXN], revin[MAXN];
pair <int, int> sparse[__lg(MAXN)][MAXN];
vector <int> adj[MAXN];
int n, m, q;
int cnt = 0;
void dfs (int pos, int par, int dep = 1) {
depth[pos] = dep;
sparse[0][++cnt] = {dep, pos};
in[pos] = cnt; revin[in[pos]] = pos;
for (auto j : adj[pos]) {
if (j != par) {
dfs(j, pos, dep + 1);
sparse[0][++cnt] = {dep, pos};
}
}
out[pos] = dep;
}
int sparse2[MAXN][__lg(MAXN)];
int query (int l, int r) {
if (l > r) swap(l, r);
int x = __lg(r - l + 1);
return min(sparse[x][l], sparse[x][r - (1 << x) + 1]).second;
}
int query2 (int l, int r) {
int x = __lg(r - l + 1);
return query(in[sparse2[x][l]], in[sparse2[x][r - (1 << x) + 1]]);
}
int arr[MAXN];
array <int, 3> queries[MAXN];
int ret[MAXN];
int ans = 0;
set <int> dd;
void add (int x) {
if (dd.empty()) {
dd.insert(in[x]);
ans = depth[x];
return;
}
auto t = dd.upper_bound(in[x]);
int r = (t == dd.end() ? -1 : *t);
int l = -1;
if (t != dd.begin()) l = *(--t);
if (r != -1) {
ans -= depth[revin[r]] - depth[query(l, r)];
ans += depth[revin[r]] - depth[query(in[x], r)];
}
if (l != -1) {
ans += depth[x] - depth[query(l, in[x])];
} else {
ans += depth[x];
}
dd.insert(in[x]);
}
void rem (int x) {
dd.erase(in[x]);
if (dd.empty()) {
ans = 0; return;
}
auto t = dd.upper_bound(in[x]);
int r = (t == dd.end() ? -1 : *t);
int l = -1;
if (t != dd.begin()) l = *(--t);
if (r != -1) {
ans += depth[revin[r]] - depth[query(l, r)];
ans -= depth[revin[r]] - depth[query(in[x], r)];
}
if (l != -1) {
ans -= depth[x] - depth[query(l, in[x])];
} else {
ans -= depth[x];
}
}
int main () {
ios::sync_with_stdio(0);
cin.tie(0);
cin >> n >> m >> q;
for (int i = 1; i < n; i++) {
int a, b;
cin >> a >> b;
adj[a].push_back(b);
adj[b].push_back(a);
}
dfs(1, -1);
for (int i = 1; i <= __lg(cnt); i++) {
for (int j = 1; j + (1 << i) - 1 <= cnt; j++) {
sparse[i][j] = min(sparse[i - 1][j], sparse[i - 1][j + (1 << (i - 1))]);
}
}
for (int i = 1; i <= m; i++) {
cin >> arr[i];
sparse2[0][i] = arr[i];
}
for (int i = 1; i <= __lg(m); i++) {
for (int j = 1; j + (1 << i) - 1 <= m; j++) {
sparse2[i][j] = query(in[sparse2[i - 1][j]], in[sparse2[i - 1][j + (1 << (i - 1))]]);
}
}
/*for (int i = 1; i <= q; i++) {
int l, r;
cin >> l >> r;
vector <int> g;
for (int j = l; j <= r; j++) g.push_back(arr[j]);
sort(g.begin(), g.end(), [&] (int a, int b) {
return in[a] < in[b];
});
int ans = depth[g[0]], lca = g[0];
for (int j = 1; j < (int)g.size(); j++) {
ans += depth[g[j]] - depth[query(in[g[j]], in[g[j - 1]])];
lca = query(in[lca], in[g[j]]);
}
ans -= depth[lca]; ans++;
cout << ans << '\n';
}*/
for (int i = 1; i <= q; i++) {
cin >> queries[i][0] >> queries[i][1]; queries[i][2] = i;
}
int x = sqrt(m); while (x * x < m) x++; while (x * x > m) x--;
sort(queries + 1, queries + q + 1, [&] (array <int, 3> &a, array <int, 3> &b) {
return a[0] / x == b[0] / x ? a[1] < b[1] : a[0] / x < b[0] / x;
});
int l = queries[1][1] + 1, r = queries[1][1];
for (int i = 1; i <= q; i++) {
int tl = queries[i][0], tr = queries[i][1];
while (l < tl) rem(arr[l++]);
while (l > tl) add(arr[--l]);
while (r < tr) add(arr[++r]);
while (r > tr) rem(arr[r--]);
ret[i] = ans - depth[query2(l, r)] + 1;
}
for (int i = 1; i <= q; i++) cout << ret[i] << '\n';
}
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
6 ms |
31068 KB |
Output is correct |
2 |
Incorrect |
4 ms |
31068 KB |
Output isn't correct |
3 |
Halted |
0 ms |
0 KB |
- |
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
6 ms |
31068 KB |
Output is correct |
2 |
Incorrect |
4 ms |
31068 KB |
Output isn't correct |
3 |
Halted |
0 ms |
0 KB |
- |
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Incorrect |
5 ms |
31068 KB |
Output isn't correct |
2 |
Halted |
0 ms |
0 KB |
- |
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
4 ms |
31068 KB |
Output is correct |
2 |
Incorrect |
51 ms |
74012 KB |
Output isn't correct |
3 |
Halted |
0 ms |
0 KB |
- |
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
5 ms |
31064 KB |
Output is correct |
2 |
Incorrect |
4 ms |
24924 KB |
Output isn't correct |
3 |
Halted |
0 ms |
0 KB |
- |
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
6 ms |
31068 KB |
Output is correct |
2 |
Incorrect |
4 ms |
31068 KB |
Output isn't correct |
3 |
Halted |
0 ms |
0 KB |
- |