#include <bits/stdc++.h>
#define ll long long
#define pb push_back
#define eb emplace_back
#define pu push
#define ins insert
#define fi first
#define se second
#define all(a) a.begin(),a.end()
#define bruh ios_base::sync_with_stdio(false);cin.tie(0);cout.tie(0);
#define fu(x,a,b) for (auto x=a;x<=b;x++)
#define fd(x,a,b) for (auto x=a;x>=b;x--)
#define int ll
using namespace std;
//mt19937 mt(chrono::steady_clock::now().time_since_epoch().count());
/*
Competitive Programming notes that I need to study & fix my dumbass self:
1. Coding:
- Always be sure to check the memory of arrays (maybe use vectors), for loops
- Always try to maximize the memory if possible, even if you are going for subtasks
- Do not exploit #define int long long, it will kill you
2. Stress: 
- Always try generating big testcases and try if they run
3. Time management:
- Don't overcommit or undercommit, always spend a certain amount of time to think a problem, don't just look at it and say I'm fucked
- Do not spend too much time coding brute-force solutions, they should be easily-codable solutions that don't take up too much time
Time management schedule:
Offline / LAH days (4 problems - 3h):
15' thinking of solution / idea
1. no idea: skip
2. yes idea: continue thinking for <= 15'
+ implementing: <= 20'
+ brute-force: <= 5'
+ test generator: <= 5'
I hate offline because I am dumb
*/
typedef pair<int, int> ii;
const int N = 2e5+5;
const int M = 25;
const int B = 750;
const int mod = 1e9+7;
const int inf = 1e18;
using cd = complex<double>;
const long double PI = acos(-1);
int power(int a,int b) {ll x = 1;if (a >= mod) a%=mod; while (b) {if (b & 1) x = x*a % mod;a = a*a % mod;b>>=1;}return x;} 
int n,m;
int a[N][M], cnt[M];
int val[N];
int dp[2000005][2];
int rev(int mask) 
{
	return ~mask & ((1<<m) - 1);
}
int calc(int i, int mask) 
{
	return __builtin_popcount(mask & val[i]);
}
void solve()
{
	cin>>n>>m;
	for (int i = 0; i < (1<<m); i++) dp[i][0] = dp[i][1] = 0;
	for (int i = 1; i <= n; i++) 
	{
		for (int j = 0; j < m; j++) 
		{
			cin>>a[i][j];
			if (!a[i][j]) val[i] |= (1<<j);
			cnt[j] += a[i][j];
		}
		// cout<<val[i]<<" ";
		if (calc(i, val[i]) > calc(dp[val[i]][0], val[i])) dp[val[i]][1] = dp[val[i]][0], dp[val[i]][0] = i;
		else if (calc(i, val[i]) > calc(dp[val[i]][1], val[i])) dp[val[i]][1] = i;
	}
	for (int i = 0; i < m; i++) 
	{
		for (int mask = (1<<m)-1; mask >= 0; mask--) 
		{
			if ((mask>>i & 1) == 0) 
			{
				int nmask = mask ^ (1<<i);
				if (dp[nmask][0] != dp[mask][0] && dp[nmask][0] != dp[mask][1])
				{
					int v1 = calc(dp[nmask][0], mask);
					if (v1 > calc(dp[mask][0], mask)) 
					{
						dp[mask][1] = dp[mask][0];
						dp[mask][0] = dp[nmask][0];
					} else if (v1 > calc(dp[mask][1], mask))
					{
						dp[mask][1] = dp[nmask][0];
					} 	
				}
				if (dp[nmask][1] != dp[mask][0] && dp[nmask][1] != dp[mask][1]) 
				{
					int v2 = calc(dp[nmask][1], mask);
					if (v2 > calc(dp[mask][0], mask)) 
					{
						dp[mask][1] = dp[mask][0];
						dp[mask][0] = dp[nmask][1];
					} else if (v2 > calc(dp[mask][1], mask)) dp[mask][1] = dp[nmask][1]; 
				}
			}
		}
	}
	for (int i = 0; i < m; i++) 
	{
		for (int mask = 0; mask < (1<<m); mask++) 
		{
			if (mask>>i & 1) 
			{
				int nmask = mask ^ (1<<i);
				if (dp[nmask][0] != dp[mask][0] && dp[nmask][0] != dp[mask][1])
				{
					int v1 = calc(dp[nmask][0], mask);
					if (v1 > calc(dp[mask][0], mask)) 
					{
						dp[mask][1] = dp[mask][0];
						dp[mask][0] = dp[nmask][0];
					} else if (v1 > calc(dp[mask][1], mask))
					{
						dp[mask][1] = dp[nmask][0];
					} 	
				}
				if (dp[nmask][1] != dp[mask][0] && dp[nmask][1] != dp[mask][1]) 
				{
					int v2 = calc(dp[nmask][1], mask);
					if (v2 > calc(dp[mask][0], mask)) 
					{
						dp[mask][1] = dp[mask][0];
						dp[mask][0] = dp[nmask][1];
					} else if (v2 > calc(dp[mask][1], mask)) dp[mask][1] = dp[nmask][1]; 
				}
			}
		}
	}
	// cout<<dp[3][0].fi<<" "<<dp[3][0].se<<" "<<dp[3][1].fi<<" "<<dp[3][1].se<<endl;
	// cout<<endl;
	int bound = n / 2;
	for (int i = 1; i <= n; i++) 
	{
		int mask = 0, ans = 0, c = 0;
		for (int j = 0; j < m; j++) 
		{
			if (cnt[j] - a[i][j] > bound) ans++;
			else if (cnt[j] - a[i][j] == bound) 
			{
				mask |= (1<<j);
			}
		}
		// cout<<endl;
		// cout<<dp[mask][0].fi<<" "<<dp[mask][0].se<<" "<<dp[mask][1].fi<<" "<<dp[mask][1].se<<" "<<endl;
		if (dp[mask][0] != i) cout<<ans + __builtin_popcount(mask & val[dp[mask][0]])<<endl;
		else cout<<ans + __builtin_popcount(mask & val[dp[mask][1]])<<endl;
	}
}
/*
Go through the mistakes you usually make and revise your code, for god's sake...
*/
signed main()
{
	bruh
	//freopen("input.inp","r",stdin);
	//freopen("output.inp","w",stdout);
	int t = 1;
	// cin>>t;
	while (t--)
	{
		solve();
		cout<<"\n";
	}
}
| # | Verdict | Execution time | Memory | Grader output | 
|---|
| Fetching results... | 
| # | Verdict | Execution time | Memory | Grader output | 
|---|
| Fetching results... | 
| # | Verdict | Execution time | Memory | Grader output | 
|---|
| Fetching results... | 
| # | Verdict | Execution time | Memory | Grader output | 
|---|
| Fetching results... | 
| # | Verdict | Execution time | Memory | Grader output | 
|---|
| Fetching results... | 
| # | Verdict | Execution time | Memory | Grader output | 
|---|
| Fetching results... | 
| # | Verdict | Execution time | Memory | Grader output | 
|---|
| Fetching results... |