#include <algorithm>
#include <atcoder/all>
#include <bitset>
#include <cassert>
#include <cmath>
#include <ctime>
#include <functional>
#include <iomanip>
#include <iostream>
#include <map>
#include <queue>
#include <random>
#include <set>
#include <stack>
#include <string>
#include <tuple>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
using namespace std;
using namespace atcoder;
typedef long long ll;
#define rep(i, n) for (int i = 0; i < (int)(n); i++)
#define repr(i, n) for (int i = (int)(n) - 1; i >= 0; i--)
#define repk(i, k, n) for (int i = k; i < (int)(n); i++)
#define all(v) v.begin(), v.end()
#define mod1 1000000007
#define mod2 998244353
#define mod3 100000007
#define vi vector<int>
#define vs vector<string>
#define vc vector<char>
#define vl vector<ll>
#define vb vector<bool>
#define vvi vector<vector<int>>
#define vvc vector<vector<char>>
#define vvl vector<vector<ll>>
#define vvb vector<vector<bool>>
#define vvvi vector<vector<vector<int>>>
#define vvvl vector<vector<vector<ll>>>
#define pii pair<int, int>
#define pil pair<int, ll>
#define pli pair<ll, int>
#define pll pair<ll, ll>
#define vpii vector<pair<int, int>>
#define vpll vector<pair<ll, ll>>
#define vvpii vector<vector<pair<int, int>>>
#define vvpll vector<vector<pair<ll, ll>>>

template <typename T>
void debug(T e) {
    cerr << e << endl;
}

template <typename T>
void debug(vector<T> &v) {
    rep(i, v.size()) { cerr << v[i] << " "; }
    cerr << endl;
}

template <typename T>
void debug(vector<vector<T>> &v) {
    rep(i, v.size()) {
        rep(j, v[i].size()) { cerr << v[i][j] << " "; }
        cerr << endl;
    }
}

template <typename T>
void debug(vector<pair<T, T>> &v) {
    rep(i, v.size()) { cerr << v[i].first << " " << v[i].second << endl; }
}

template <typename T>
void debug(set<T> &st) {
    for (auto itr = st.begin(); itr != st.end(); itr++) {
        cerr << *itr << " ";
    }
    cerr << endl;
}

template <typename T>
void debug(multiset<T> &ms) {
    for (auto itr = ms.begin(); itr != ms.end(); itr++) {
        cerr << *itr << " ";
    }
    cerr << endl;
}

template <typename T>
void debug(map<T, T> &mp) {
    for (auto itr = mp.begin(); itr != mp.end(); itr++) {
        cerr << itr->first << " " << itr->second << endl;
    }
}

void debug_out() { cerr << endl; }

template <typename Head, typename... Tail>
void debug_out(Head H, Tail... T) {
    cerr << H << " ";
    debug_out(T...);
}

using mint = modint998244353;

void debug_mint1(vector<mint> &vec) {
    for (int i = 0; i < vec.size(); i++) {
        cerr << vec[i].val() << " ";
    }
    cerr << endl;
}

void debug_mint2(vector<vector<mint>> &vec) {
    for (int i = 0; i < vec.size(); i++) {
        for (int j = 0; j < vec[i].size(); j++) {
            cerr << vec[i][j].val() << " ";
        }
        cerr << endl;
    }
}


int main() {
    ll N, M;
    cin >> N >> M;
    vector<vector<ll>> A(N, vector<ll>(M));
    for (ll i = 0; i < N; i++){
        for (ll j = 0; j < M; j++){
            cin >> A[i][j];
        }
    }

    ll half = N / 2;
    
    if (M == 1){
        ll ans = 0;
        for (ll i = 0; i < half; i++){
            ans -= A[i][0];
        }
        for (ll i = N - half; i < N; i++){
            ans += A[i][0];
        }
        cout << ans << endl;
        return 0;
    }
    else if (M == 2){
        ll INF = 100000000000000023;
        // 前半の最小値
        vector<vector<ll>> dp_form(half + 1, vector<ll>(half + 1, INF));
        dp_form[0][0] = 0;
        for (ll i = 0; i < half; i++){
            for (ll j = 0; j <= i; j++){
                dp_form[i + 1][j] = min(dp_form[i + 1][j], dp_form[i][j] + A[i][0]);
                dp_form[i + 1][j + 1] = min(dp_form[i + 1][j + 1], dp_form[i][j] + A[i][1]);
            }
        }

        // 後半の最大値
        vector<vector<ll>> dp_latt(half + 1, vector<ll>(half + 1, -INF));
        dp_latt[0][0] = 0;
        for (ll i = 0; i < half; i++){
            for (ll j = 0; j <= i; j++){
                dp_latt[i + 1][j] = max(dp_latt[i + 1][j], dp_latt[i][j] + A[i + N - half][0]);
                dp_latt[i + 1][j + 1] = max(dp_latt[i + 1][j + 1], dp_latt[i][j] + A[i + N - half][1]);
            }
        }
        ll ans = 0;
        for (ll i = 0; i <= half; i++){
            ans = max(ans, -dp_form[half][i] + dp_latt[half][i]);
        }
        cout << ans << endl;
        return 0;
    }
    vector<vector<ll>> mat(half, vector<ll>(half, 0));
    for (ll i = 0; i < half; i++){
        for (ll j = 0; j < half; j++){
            for (ll k = 0; k < M; k++){
                mat[i][j] = max(mat[i][j], A[N - half + j][k] - A[i][k]);
            }
        }
    }
    // debug(mat);

    ll max_mat = 0;
    for (ll i = 0; i < half; i++){
        for (ll j = 0; j < half; j++){
            max_mat = max(max_mat, mat[i][j]);
        }
    }

    ll offset = max_mat * half;
    for (ll i = 0; i < half; i++){
        for (ll j = 0; j < half; j++){
            mat[i][j] = max_mat - mat[i][j];
        }
    }

    mcf_graph<ll, ll> graph(half * 2 + 2);
    for (ll i = 0; i < half; i++){
        graph.add_edge(half * 2, i, 1, 0);
        graph.add_edge(i + half, half * 2 + 1, 1, 0);
    }
    for (ll i = 0; i < half; i++){
        for (ll j = 0; j < half; j++){
            graph.add_edge(i, j + half, 1, mat[i][j]);
        }
    }

    ll fl = graph.flow(half * 2, half * 2 + 1).second;
    cout << offset - fl << endl;
}