#include <bits/stdc++.h>
#include <ext/pb_ds/tree_policy.hpp>
#include <ext/pb_ds/assoc_container.hpp>
using namespace std;
using namespace __gnu_pbds;
typedef long long ll;
typedef long double ld;
typedef complex<ld> cd;
typedef pair<int, int> pi;
typedef pair<ll,ll> pl;
typedef pair<ld,ld> pd;
typedef vector<int> vi;
typedef vector<ld> vd;
typedef vector<ll> vl;
typedef vector<pi> vpi;
typedef vector<pl> vpl;
typedef vector<cd> vcd;
template <class T> using Tree = tree<T, null_type, less<T>, rb_tree_tag,tree_order_statistics_node_update>;
#define FOR(i, a, b) for (int i=a; i<(b); i++)
#define F0R(i, a) for (int i=0; i<(a); i++)
#define FORd(i,a,b) for (int i = (b)-1; i >= a; i--)
#define F0Rd(i,a) for (int i = (a)-1; i >= 0; i--)
#define sz(x) (int)(x).size()
#define mp make_pair
#define pb push_back
#define f first
#define s second
#define lb lower_bound
#define ub upper_bound
#define all(x) x.begin(), x.end()
const int MOD = 1000000007;
const ll INF = 1e18;
const int MX = 100001;
int N,M;
ll ans = -INF, A[1000][1000], bes[1000][1000];
ll solve() {
ll ret = -INF;
F0Rd(i,N) F0Rd(j,M) {
bes[i][j] = -INF;
if (i < N-1) bes[i][j] = max(bes[i][j],bes[i+1][j]);
if (j < M-1) bes[i][j] = max(bes[i][j],bes[i][j+1]);
bes[i][j] = max(bes[i][j],A[i][j]-i-j);
ret = max(ret,bes[i][j]-A[i][j]+i+j-1);
}
return ret;
}
int main() {
ios_base::sync_with_stdio(0); cin.tie(0);
cin >> N >> M;
F0R(i,N) F0R(j,M) cin >> A[i][j];
ans = solve();
F0R(i,N) for (int j = 0; j < M-1-j; ++j) swap(A[i][j],A[i][M-1-j]);
ans = max(ans,solve());
for (int i = 0; i < N-1-i; ++i) F0R(j,M) swap(A[i][j],A[N-1-i][j]);
ans = max(ans,solve());
F0R(i,N) for (int j = 0; j < M-1-j; ++j) swap(A[i][j],A[i][M-1-j]);
ans = max(ans,solve());
cout << ans;
}
/* Look for:
* the exact constraints (multiple sets are too slow for n=10^6 :( )
* special cases (n=1?)
* overflow (ll vs int?)
* array bounds
*/
