#include "festival.h"
#include <bits/stdc++.h>
using namespace std;
#ifndef DEBUG
#define dbg(...)
#endif
// #define cout cerr << "OUT: "
#define endl '\n'
#define endlfl '\n' << flush
#define quit(x) return (void)(cout << x << endl)
template<typename T> void read(T& x) { cin >> x; }
template<typename T1, typename T2> void read(pair<T1, T2>& p) { read(p.first), read(p.second); }
template<typename T> void read(vector<T>& v) { for (auto& x : v) read(x); }
template<typename T1, typename T2> void read(T1& x, T2& y) { read(x), read(y); }
template<typename T1, typename T2, typename T3> void read(T1& x, T2& y, T3& z) { read(x), read(y), read(z); }
template<typename T1, typename T2, typename T3, typename T4> void read(T1& x, T2& y, T3& z, T4& zz) { read(x), read(y), read(z), read(zz); }
template<typename T> void print(vector<T>& v) { for (auto& x : v) cout << x << ' '; cout << endl; }
#define sz(x) (int)(x.size())
#define all(x) (x).begin(), (x).end()
#define rall(x) (x).rbegin(), (x).rend()
#define make_unique(v) sort(all(v)), v.erase(unique(all(v)), (v).end())
#define pb push_back
#define mp(a, b) make_pair(a, b)
#define fi first
#define se second
#define cmin(a, b) a = min(a, b)
#define cmax(a, b) a = max(a, b)
#define YES cout << "YES" << endl
#define NO cout << "NO" << endl
#define QYES quit("YES")
#define QNO quit("NO")
#define int long long
#define double long double
const int inf = numeric_limits<int>::max() / 4;
mt19937 rng(chrono::steady_clock::now().time_since_epoch().count());
const int MOD = 1e9+7, LOG=20;
vector<signed> calc(vector<signed> ans, vector<vector<pair<int, int>>>& vals) {
vector<signed> res;
for (auto& x : ans) res.pb(x);
for (int i=0; i<4; i++) for (auto& [p, id] : vals[i]) res.pb(id);
return res;
}
vector<signed> max_coupons(signed AA, vector<signed> P, vector<signed> T) {
int n=sz(P), A=AA;
vector<vector<pair<int, int>>> vals(4);
for (int i=0; i<sz(P); i++) {
vals[T[i]-1].pb({P[i], i});
}
for (int i=0; i<4; i++) sort(rall(vals[i]));
vector<signed> ans;
while (n--) {
int best=inf, act=0;
vector<int> v;
for (int i=1; i<4; i++) if (!vals[i].empty() && (A-vals[i].back().fi)*(i+1)>A) {
v.pb(i);
}
if (v.empty()) break;
do {
int sum=0;
for (int i=0; i<sz(v); i++) {
int cur=vals[v[i]].back().fi;
for (int j=i; j<sz(v); j++) {
cur*=(v[j]+1);
}
sum+=cur;
}
if (sum<best) best=sum, act=v[0];
} while (next_permutation(all(v)));
A=(A-vals[act].back().fi)*(act+1);
assert(A>=0);
ans.pb(vals[act].back().se);
vals[act].pop_back();
if (A>=1e12) return calc(ans, vals);
}
for (int i=0; i<4; i++) while (!vals[i].empty() && (A-vals[i].back().fi)*(i+1)==A){
ans.pb(vals[i].back().se);
vals[i].pop_back();
}
for (auto &u: vals) reverse(all(u));
vector<int> pref0; if (!vals[0].empty()) pref0.pb(vals[0][0].fi);
for (int i=1; i<sz(vals[0]); i++) pref0.pb(pref0[i-1]+vals[0][i].fi);
int m1=min(sz(vals[1]), 40LL)+1, m2=min(sz(vals[2]), 40LL)+1, m3=min(sz(vals[3]), 40LL)+1;
vector<vector<vector<int>>> dp(m1, vector<vector<int>>(m2, vector<int>(m3, -1))); dp[0][0][0]=A;
int best=0;
vector<int> besti(3);
for (int i=0; i<m1; i++) {
for (int j=0; j<m2; j++) {
for (int k=0; k<m3; k++) {
if (i) cmax(dp[i][j][k], (dp[i-1][j][k] - vals[1][i-1].fi)*2);
if (j) cmax(dp[i][j][k], (dp[i][j-1][k] - vals[2][j-1].fi)*3);
if (k) cmax(dp[i][j][k], (dp[i][j][k-1] - vals[3][k-1].fi)*4);
if (dp[i][j][k]<0) continue;
auto lb=upper_bound(all(pref0), dp[i][j][k]) - pref0.begin();
if (i+j+k+lb>best) {
best=i+j+k+lb;
besti={i, j, k};
dbg(i, j, k, lb);
}
}
}
}
vector<int> vec(1, 0);
if (best==0) return ans;
vector<signed> res;
for (int i=0; i<best - besti[0] - besti[1] - besti[2]; i++) res.pb(vals[0][i].se);
reverse(all(res));
best=besti[0]+besti[1]+besti[2];
while (best--) {
int i=besti[0], j=besti[1], k=besti[2];
if (i && dp[i][j][k]==(dp[i-1][j][k] - vals[1][i-1].fi)*2) {
res.pb(vals[1][i-1].se);
besti[0]--;
} else if (j && dp[i][j][k]==(dp[i][j-1][k] - vals[2][j-1].fi)*3) {
res.pb(vals[2][j-1].se);
besti[1]--;
} else {
res.pb(vals[3][k-1].se);
besti[2]--;
}
}
reverse(all(res));
ans.insert(ans.end(), all(res));
return ans;
}
