# include <bits/stdc++.h>
using namespace std;
using ll = long long;
using ull = unsigned long long;
const double pi = acos(-1);
template<class T>constexpr T inf() { return ::std::numeric_limits<T>::max(); }
template<class T>constexpr T hinf() { return inf<T>() / 2; }
template <typename T_char>T_char TL(T_char cX) { return tolower(cX); }
template <typename T_char>T_char TU(T_char cX) { return toupper(cX); }
template<class T> bool chmin(T& a,T b) { if(a > b){a = b; return true;} return false; }
template<class T> bool chmax(T& a,T b) { if(a < b){a = b; return true;} return false; }
template<class T> bool is_sqare(T a) { if(floor(sqrt(a)) * floor(sqrt(a)) == a){ return true; }return false; }
int popcnt(unsigned long long n) { int cnt = 0; for (int i = 0; i < 64; i++)if ((n >> i) & 1)cnt++; return cnt; }
int d_sum(ll n) { int ret = 0; while (n > 0) { ret += n % 10; n /= 10; }return ret; }
int d_cnt(ll n) { int ret = 0; while (n > 0) { ret++; n /= 10; }return ret; }
ll gcd(ll a, ll b) { if (b == 0)return a; return gcd(b, a%b); };
ll lcm(ll a, ll b) { ll g = gcd(a, b); return a / g*b; };
template<class T> using dijk = priority_queue<T, vector<T>, greater<T>>;
# define all(qpqpq)           (qpqpq).begin(),(qpqpq).end()
# define UNIQUE(wpwpw)        sort(all((wpwpw)));(wpwpw).erase(unique(all((wpwpw))),(wpwpw).end())
# define LOWER(epepe)         transform(all((epepe)),(epepe).begin(),TL<char>)
# define UPPER(rprpr)         transform(all((rprpr)),(rprpr).begin(),TU<char>)
# define rep(i,upupu)         for(ll i = 0, i##_len = (upupu);(i) < (i##_len);(i)++)
# define reps(i,opopo)        for(ll i = 1, i##_len = (opopo);(i) <= (i##_len);(i)++)
# define len(x)                ((int)(x).size())
# define bit(n)               (1LL << (n))
# define pb push_back
#ifdef LOCAL
#  include "_debug_print.hpp"
#  define debug(...) debug_print::multi_print(#__VA_ARGS__, __VA_ARGS__)
#else
#  define debug(...) (static_cast<void>(0))
#endif

struct INIT{
    INIT(){
        std::ios::sync_with_stdio(false);
        std::cin.tie(0);
        cout << fixed << setprecision(20);
    }
}INIT;

void solve(){
    int h, w;
    cin >> h >> w;
    vector<vector<int>> a(h, vector<int>(w));
    rep(i, h)rep(j, w)cin >> a[i][j];
    ll ans = 0;
    const int dy[8] = {-1, -1, 0, 1, 1, 1, 0, -1};
    const int dx[8] = {0, 1, 1, 1, 0, -1, -1, -1};
    auto within = [&](int y, int x) -> bool {
        return 0 <= y && y < h && 0 <= x && x < w;
    };
    map<tuple<int, int, int>, ll> mp;
    vector<vector<vector<ll>>> v(h, vector<vector<ll>>(w, vector<ll>(8, 0)));
    rep(i, h)rep(j, w){
        rep(dir1, 8){
            ll cur = 0;
            {
                int y = i, x = j;
                while(within(y, x)){
                    cur += a[y][x];
                    y += dy[dir1];
                    x += dx[dir1];
                }
            }
            v[i][j][dir1] = cur;
        }
    }
    rep(i, h)rep(j, w){
        rep(dir1, 8)rep(dir2, 8){
            ll cur = -a[i][j];
            if(dir1 == dir2)continue;
            if((dir1 - dir2 + 8) % 8 == 4)continue;
            cur += v[i][j][dir1];
            cur += v[i][j][(dir1 + 4) % 8];
            cur -= a[i][j];
            cur += v[i][j][dir2];
            cur += v[i][j][(dir2 + 4) % 8];
            cur -= a[i][j];
            chmax(ans, cur);
        }
    }
    debug(ans);
    rep(i, h)rep(i2, h){
        if(i == i2)continue;
        chmax(ans, v[i][0][2] + v[i2][0][2]);
    }
    debug(ans);
    rep(j, w)rep(j2, w){
        if(j == j2)continue;
        chmax(ans, v[0][j][4] + v[0][j2][4]);
    }
    debug(ans);
    rep(x, h + w - 1)rep(x2, h + w - 1){
        if(x == x2)continue;
        ll cur = 0;
        if(x < h){
            cur += v[x][0][1];
        }else{
            cur += v[h - 1][x - h + 1][1];
        }

        if(x2 < h){
            cur += v[x2][0][1];
        }else{
            cur += v[h - 1][x2 - h + 1][1];
        }
        chmax(ans, cur);
    }
    debug(ans);
    rep(x, h + w - 1)rep(x2, h + w - 1){
        if(x == x2)continue;
        ll cur = 0;
        if(x < w){
            cur += v[0][x][3];
        }else{
            cur += v[x - w + 1][0][3];
        }

        if(x2 < w){
            cur += v[0][x2][3];
        }else{
            cur += v[x2 - w + 1][0][3];
        }
        chmax(ans, cur);
    }
    debug(ans);
    cout << ans << endl;
}

int main(){
    int t = 1;
    //cin >> t;
    while(t--)solve();
}