为什么我的代码会超时，好像和标准代码复杂度是一样的

#include <bits/stdc++.h>
#include<vector>
#include<iostream>
#define ll long long
#define int long long
#define ull unsigned long long
#define vb v.begin()
#define ve v.end()
#define vc v.clear()
#define vs (int)v.size()
#define ss (int)s.size()
#define rs(i) resize(i)
#define ft first
#define sd second
#define inf 0x3f3f3f3f
#define P pair<int, int>
#define Pll pair<ll, ll>
#define Pis pair<int, string>
#define Psi pair<string, int>
#define Pic pair<int, char>
#define Pci pair<char, int>
#define forn(i, a, b) for (int i = a; i < b; i++)
#define fore(i, a, b) for (int i = a; i <= b; i++)
#define rofn(i, a, b) for (int i = a; i > b; i--)
#define rofe(i, a, b) for (int i = a; i >= b; i--)
#define Ios ios::sync_with_stdio(false), cin.tie(0)
const ll N = 200010;
const ll M = 10010;
const int INF = 2147483647;
using namespace std;
int u[N], s[N];
map<int, vector<int>> q, SUM;
void solve()
{
	q.clear();
	SUM.clear();
	int n;
	cin >> n;
	int sum = 0;
	fore(i, 1, n) cin >> u[i];
	fore(i, 1, n) cin >> s[i];
	fore(i, 1, n)
	{
		q[u[i]].push_back(s[i]);
		sum += s[i];
	}
	for (auto it : q)
	{
		sort(q[it.first].begin(), q[it.first].end());
	}
	for (auto it : q)
	{
		SUM[it.first].push_back(0);
		fore(i, 1, it.second.size())
		{
			SUM[it.first].push_back(q[it.first][i-1]);
			SUM[it.first][i] += SUM[it.first][i - 1] ;
		}
		fore(i, it.second.size(),n) SUM[it.first].push_back(SUM[it.first][i-1]);
	}
	fore(k, 1, n)
	{
		int tmp = sum;
		for (auto it : q)
		{
			int out = it.second.size() % k;
			tmp -= SUM[it.first][out];
			/*fore(i, 1, out)
			{
				tmp -= it.second[i - 1];
			}*/
		}
		cout << tmp << ' ';
	}
	cout << endl;
}
signed main()
{
	Ios;
	int t;
	cin >> t;
	while (t--)
	{
		solve();
	}
	return 0;
}
/*

*/
/*******************************************
				  坤坤助我
............................................
............................................
............................................
............................................
.............................1ffLf1.........
............................GGGGGGGC........
....................8CCCftCCCCC1;LLG,.......
..................:CC8CCC.CCCCCCii..........
.................CC88CCCC.CCCCCCC...........
................CC88CCCCC .8.:8CCC..........
................888 CCCCC.88888G;;,.........
,................888CC88.888888888,.........
:::...............CCC888.888888888..........
:::::............Gi1tC8f888888C88,..........
:::::::,........111tttfffffLG8888...........
:::::::::,....ii111ttttttfffttG.............
::::::::::::i111tttffffffffffttt............
::::::::::11tttttfff....fffffftt ...........
..:::::::1Lf1tff,..........fffff1...........
....,:::iiftG::::::.........,LLGtL..........
::::::;:i11t:;;;;;;::,,,,,,,,, f1ii,::,,,,,,
::::::::i11t:::::::::::::::::::f111:::::::::
::::::::111:::::::::::::::::::::t111::::::::
:::::::i1if::::::,,,,,,,,,,,,,,:i111::::::,:
,,,,,::,8G,,,,,,,,,,,,,,,,,,,,,,,tL1:,,,,,,,
,,,,,,:LGL;,,,,,,,,,,,,,,,,,,,,,,,CLf,,,,,,,
,,,,,;LGGG,,,,:,,,,,,,,,,,,,,,,,,,LGCG,,,,,,

*******************************************/

我的

#include <bits/stdc++.h>

#define forn(i, n) for (int i = 0; i < int(n); i++)

using namespace std;


int main() {
    int t;
    scanf("%d", &t);
    forn(_, t){
        int n;
        scanf("%d", &n);
        vector<int> s(n), u(n);
        forn(i, n){
            scanf("%d", &s[i]);
            --s[i];
        }
        forn(i, n){
            scanf("%d", &u[i]);
        }
        vector<vector<int>> bst(n);
        forn(i, n) bst[s[i]].push_back(u[i]);
        forn(i, n) sort(bst[i].begin(), bst[i].end(), greater<int>());
        vector<vector<long long>> pr(n, vector<long long>(1, 0));
        forn(i, n) for (int x : bst[i]) pr[i].push_back(pr[i].back() + x);
        vector<long long> ans(n);
        forn(i, n) for (int k = 1; k <= int(bst[i].size()); ++k)
            ans[k - 1] += pr[i][bst[i].size() / k * k];
        forn(i, n)
            printf("%lld ", ans[i]);
        puts("");
    }
    return 0;
}

std