#include "bits/stdc++.h"
#include <unordered_map>
#include <unordered_set>

#include <stdio.h>
#include <math.h>
#include <cassert>

#include "atcoder/all"
#include <float.h>

//#include <boost/multiprecision/cpp_int.hpp>
//using namespace boost::multiprecision;
// std::fixed << std::setprecision(10) <<

using namespace std;
using namespace atcoder;

using ll = long long;
using ull = unsigned long long;
using mint = modint998244353;
using mint2 = modint1000000007;

const double eps = 1e-9;

#define REP(i, n) for (ll i = 0; i < ll(n); ++i)
#define REPI(i, init, n) for (ll i = init; i < ll(n); ++i)
#define REPD(i, init) for (ll i = init; i >=0; --i)
#define REPDI(i, init, n) for (ll i = init; i >=n; --i)
#define REPE(i, c) for (const auto& i : c)
#define TCASE() ll _ttt; cin >> _ttt; while(_ttt--)

using vl = vector<ll>;
#define VL(a, n) vector<ll> a; a.assign(n, 0);
#define VLI(a, n) vector<ll> a; a.assign(n, 0); for(auto& x : a) cin >> x;
#define VSI(a, n) vector<string> a; a.assign(n, ""); for(auto& x : a) cin >> x;
using vvl = vector<vector<ll>>;
#define VVLI(a, n, m, init) vector<vector<ll>> a(n); for(auto& x : a) x.assign(m, init);
using vd = vector<double>;
using pl = pair<ll, ll>;

struct uv { ll u; ll v; ll c; };
#define VUVI(a, n) vector<uv> a; a.assign(n, {0, 0, 1}); for(auto& x : a) {cin >> x.u >> x.v; x.u--; x.v--;}
#define VUVCI(a, n) vector<uv> a; a.assign(n, {0, 0, 0}); for(auto& x : a) {cin >> x.u >> x.v >> x.c; x.u--; x.v--;}
vvl to_edge(const ll n, const vector<uv>& v) { vvl ret(n); for (auto& x : v) ret[x.u].push_back(x.v); return ret; }
vvl to_edge_d(const ll n, const vector<uv>& v) { vvl ret(n); for (auto& x : v) { ret[x.u].push_back(x.v); ret[x.v].push_back(x.u); } return ret; }

template <class T = long long> using pql = priority_queue<T>;
template <class T = long long> using pqg = priority_queue<T, vector<T>, greater<T>>;
using vm = vector<mint>;
using vvm = vector<vm>;

void yn(bool f) { std::cout << (f ? "Yes" : "No") << endl; };
template<class T> void ov(const T& v) { for (auto it = v.begin(); it != v.end(); it++) { if (it != v.begin()) cout << " "; cout << *it; } };
template<> void ov(const vm& v) { for (auto it = v.begin(); it != v.end(); it++) { if (it != v.begin()) cout << " "; cout << it->val(); } };

const ll llhuge = 1LL << 60; //十分でかいが多少足しても溢れない数値
auto gmax(const auto& a, const auto& b) { return a > b ? a : b; };
auto gmin(const auto& a, const auto& b) { return a < b ? a : b; };
template<class T, class U> bool chmin(T& a, const U& b) { const T olda = a; a = gmin(a, b); return olda != a; }
template<class T, class U> bool chmax(T& a, const U& b) { const T olda = a; a = gmax(a, b); return olda != a; }
template<class T> void mysort(T& v) { std::sort(begin(v), end(v)); };
template<class T, class U> void mysort(T& v, U pr) { std::sort(begin(v), end(v), pr); };
template<class T> void myrev(T& v) { std::reverse(begin(v), end(v)); };

int main() {
	cin.tie(nullptr);
	ios::sync_with_stdio(false);

	ll h, w;
	cin >> h >> w;

	vvl a(h, vl(w, 0));
	REP(i, h) REP(j, w) cin >> a[i][j];

	vl rsum(h, 0);
	vl csum(w, 0);
	vl xsum1(h + w, 0);
	vl xsum2(h + w, 0);
	REP(i, h) REP(j, w) {
		rsum[i] += a[i][j];
		csum[j] += a[i][j];
		xsum1[i + j] += a[i][j];
		xsum2[i - j + w - 1] += a[i][j];
	}

	ll ans = 0;
	REP(i1, h) {
		ll tans = rsum[i1];

		REP(i2, h) {
			if (i2 == i1) continue;
			chmax(ans, tans + rsum[i2]);
		}
		REP(i2, w) {
			chmax(ans, tans + csum[i2] - a[i1][i2]);
		}
		REP(i2, h + w) {
			if (i2 >= i1 && i2 - i1 < w) {
				chmax(ans, tans + xsum1[i2] - a[i1][i2 - i1]);
			}
			else {
				chmax(ans, tans + xsum1[i2]);
			}
		}
		REP(i2, h + w) {
			ll ti = w - (i2 - i1) - 1;
			if (ti >= 0 && ti < w) {
				chmax(ans, tans + xsum2[i2] - a[i1][ti]);
			}
			else {
				chmax(ans, tans + xsum2[i2]);
			}
		}
	}

	REP(i1, w) {
		ll tans = csum[i1];

		REP(i2, w) {
			if (i2 == i1) continue;
			chmax(ans, tans + csum[i2]);
		}
		REP(i2, h + w) {
			if (i2 >= i1 && i2 - i1 < h) {
				chmax(ans, tans + xsum1[i2] - a[i2 - i1][i1]);
			}
			else {
				chmax(ans, tans + xsum1[i2]);
			}
		}
		REP(i2, h + w) {
			ll ti = i2 - (w - 1) + i1;
			if (ti >= 0 && ti < h) {
				chmax(ans, tans + xsum2[i2] - a[ti][i1]);
			}
			else {
				chmax(ans, tans + xsum2[i2]);
			}
		}
	}
	REP(i1, h + w) {
		ll tans = xsum1[i1];

		REP(i2, h + w) {
			if (i2 == i1) continue;
			chmax(ans, tans + xsum1[i2]);
		}

		REP(i2, h + w) {
			ll tr = (i1 + i2 - w + 1) / 2;
			ll tc = (i2 - i1 + w - 1) / 2;
			if (tr >= 0 && tc >= 0 && tr < h && tc < w) {
				chmax(ans, tans + xsum2[i2] - a[tr][tc]);
			}
			else {
				chmax(ans, tans + xsum2[i2]);
			}
		}
	}
	REP(i1, h + w) {
		ll tans = xsum2[i1];

		REP(i2, h + w) {
			if (i2 == i1) continue;
			chmax(ans, tans + xsum2[i2]);
		}
	}

	cout << ans << endl;
}