// Judges with GCC >= 12 only needs Ofast
// #pragma GCC optimize("O3,no-stack-protector,fast-math,unroll-loops,tree-vectorize")
// MLE optimization
// #pragma GCC optimize("conserve-stack")
// Old judges
// #pragma GCC target("sse4.2,popcnt,lzcnt,abm,mmx,fma,bmi,bmi2")
// New judges. Test with assert(__builtin_cpu_supports("avx2"));
// #pragma GCC target("avx2,popcnt,lzcnt,abm,bmi,bmi2,fma,tune=native")
// Atcoder
// #pragma GCC target("avx2,popcnt,lzcnt,abm,bmi,bmi2,fma")
/*
#include <ext/pb_ds/assoc_container.hpp>
#include <ext/pb_ds/tree_policy.hpp>
using namespace __gnu_pbds;
typedef tree<int,null_type,less<int>,rb_tree_tag,tree_order_statistics_node_update> ordered_set;
- insert(x),erase(x)
- find_by_order(k): return iterator to the k-th smallest element
- order_of_key(x): the number of elements that are strictly smaller
*/
#include<bits/stdc++.h>
using namespace std;
mt19937_64 rng(chrono::steady_clock::now().time_since_epoch().count());
uniform_real_distribution<> pp(0.0,1.0);
#define ll long long
#define ld long double
#define pii pair<int,int>
#define piii pair<int,pii>
#define mpp make_pair
#define fi first
#define se second
const long long inf=1e18;
const int mod=998244353;
const int maxn=200005;
const int B=450;
const int maxs=655;
const int maxm=200005;
const int maxq=1000005;
const int maxl=25;
const int maxa=1000000;
const int root=3;
const int base=101;
int n,rr,q;
int c[maxn],L[maxn],R[maxn],T;
vector<int> edge[maxn];
vector<pii> qq[maxn];
vector<int> pos[maxn];
vector<int> res1[maxn],res2[maxn];
int num[maxn];
void dfs(int u){
L[u]=++T;
pos[c[u]].push_back(L[u]);
for(int v:edge[u]) dfs(v);
R[u]=T;
qq[c[u]].push_back({L[u],R[u]});
//cout << u << ' ' << c[u] << ' ' << L[u] << ' ' << R[u] << '\n';
}
void dfs2(int u,int C,int d){
res1[C][c[u]]+=d;
d+=(num[u]=(c[u]==C));
for(int v:edge[u]){
dfs2(v,C,d);
num[u]+=num[v];
}
res2[C][c[u]]+=num[u];
}
void solve(){
cin >> n >> rr >> q;
for(int i=1;i<=n;i++){
if(i>1){
int p;cin >> p;
edge[p].push_back(i);
}
cin >> c[i];
}
dfs(1);
for(int i=1;i<=rr;i++){
if((int)pos[i].size()>B){
res1[i].assign(rr+1,0);
res2[i].assign(rr+1,0);
dfs2(1,i,0);
}
}
for(int i=1;i<=q;i++){
int x,y;cin >> x >> y;
if((int)pos[x].size()>B) cout << res1[x][y] << endl;
else if((int)pos[y].size()>B) cout << res2[y][x] << endl;
else{
int total=0,j=0;
vector<pii> cc;
//cout << '*' << x << '\n';
for(auto [l,r]:qq[x]){
//cout << l << ' ' << r << '\n';
while(j<(int)pos[y].size() && pos[y][j]<=r) cc.push_back({pos[y][j++],1});
while((int)cc.size()>=2 && cc.end()[-2].fi>=l){
int x=cc.back().se;cc.pop_back();
cc.back().se+=x;
}
if(!cc.empty() && cc.back().fi>=l) total+=cc.back().se;
}
cout << total << endl;
}
}
}
signed main(){
ios_base::sync_with_stdio(false);
cin.tie(NULL);cout.tie(NULL);
int test=1;//cin >> test;
while(test--) solve();
}
# |
Verdict |
Execution time |
Memory |
Grader output |
1 |
Correct |
8 ms |
24408 KB |
Output is correct |
2 |
Correct |
10 ms |
24420 KB |
Output is correct |
3 |
Correct |
7 ms |
24920 KB |
Output is correct |
4 |
Correct |
9 ms |
25172 KB |
Output is correct |
5 |
Correct |
15 ms |
23896 KB |
Output is correct |
6 |
Correct |
21 ms |
23896 KB |
Output is correct |
7 |
Correct |
26 ms |
23892 KB |
Output is correct |
8 |
Correct |
31 ms |
23896 KB |
Output is correct |
9 |
Correct |
43 ms |
24408 KB |
Output is correct |
10 |
Correct |
55 ms |
24384 KB |
Output is correct |
11 |
Correct |
70 ms |
25176 KB |
Output is correct |
12 |
Correct |
86 ms |
26456 KB |
Output is correct |
13 |
Correct |
95 ms |
26200 KB |
Output is correct |
14 |
Correct |
137 ms |
26812 KB |
Output is correct |
15 |
Correct |
154 ms |
29988 KB |
Output is correct |
# |
Verdict |
Execution time |
Memory |
Grader output |
1 |
Correct |
566 ms |
30504 KB |
Output is correct |
2 |
Correct |
705 ms |
29028 KB |
Output is correct |
3 |
Correct |
1091 ms |
32960 KB |
Output is correct |
4 |
Correct |
161 ms |
26712 KB |
Output is correct |
5 |
Correct |
206 ms |
28684 KB |
Output is correct |
6 |
Correct |
402 ms |
31444 KB |
Output is correct |
7 |
Correct |
744 ms |
30788 KB |
Output is correct |
8 |
Correct |
602 ms |
43968 KB |
Output is correct |
9 |
Correct |
1174 ms |
35264 KB |
Output is correct |
10 |
Correct |
1633 ms |
53036 KB |
Output is correct |
11 |
Correct |
1853 ms |
35200 KB |
Output is correct |
12 |
Correct |
654 ms |
36872 KB |
Output is correct |
13 |
Correct |
965 ms |
37344 KB |
Output is correct |
14 |
Correct |
1174 ms |
39956 KB |
Output is correct |
15 |
Correct |
1535 ms |
43088 KB |
Output is correct |
16 |
Correct |
1557 ms |
52088 KB |
Output is correct |
17 |
Correct |
1601 ms |
53616 KB |
Output is correct |