#include <bits/stdc++.h>
using namespace std;
using ll = long long;
using ld = long double;
// --------------------------------------------------------
#define FOR(i,l,r) for (ll i = (l); i < (r); ++i)
#define RFOR(i,l,r) for (ll i = (r)-1; (l) <= i; --i)
#define REP(i,n) FOR(i,0,n)
#define RREP(i,n) RFOR(i,0,n)
#define ALL(c) (c).begin(), (c).end()
#define RALL(c) (c).rbegin(), (c).rend()
#define SORT(c) sort(ALL(c))
#define RSORT(c) sort(RALL(c))
#define MIN(c) *min_element(ALL(c))
#define MAX(c) *max_element(ALL(c))
#define COUNT(c,v) count(ALL(c),(v))
#define SZ(c) ((ll)(c).size())
#define BIT(b,i) (((b)>>(i)) & 1)
#define PCNT(b) __builtin_popcountll(b)
#define P0(i) (((i) & 1) == 0)
#define P1(i) (((i) & 1) == 1)
#define LB(c,v) distance((c).begin(), lower_bound(ALL(c), (v)))
#define UB(c,v) distance((c).begin(), upper_bound(ALL(c), (v)))
#define UQ(c) SORT(c), (c).erase(unique(ALL(c)), (c).end())
#define elif else if
#ifdef _LOCAL
    #define debug_bar cerr << "--------------------\n";
    #define debug(x) cerr << "l." << __LINE__ << " : " << #x << " = " << (x) << '\n'
    #define debug_pair(x) cerr << "l." << __LINE__ << " : " << #x << " = (" << x.first << "," << x.second << ")\n";
    template<class T> void debug_line(const vector<T>& ans, int l, int r, int L = 0) { cerr << "l." << L << " :"; for (int i = l; i < r; i++) { cerr << ' ' << ans[i]; } cerr << '\n'; }
#else
    #define cerr if (false) cerr
    #define debug_bar
    #define debug(x)
    #define debug_pair(x)
    template<class T> void debug_line([[maybe_unused]] const vector<T>& ans, [[maybe_unused]] int l, [[maybe_unused]] int r, [[maybe_unused]] int L = 0) {}
#endif
template<class... T> void input(T&... a) { (cin >> ... >> a); }
void print() { cout << '\n'; }
template<class T> void print(const T& a) { cout << a << '\n'; }
template<class T, class... Ts> void print(const T& a, const Ts&... b) { cout << a; (cout << ... << (cout << ' ', b)); cout << '\n'; }
template<class T> void cout_line(const vector<T>& ans, int l, int r) { for (int i = l; i < r; i++) { if (i != l) { cout << ' '; } cout << ans[i]; } cout << '\n'; }
template<class T> bool chmin(T& a, const T b) { if (b < a) { a = b; return 1; } return 0; }
template<class T> bool chmax(T& a, const T b) { if (a < b) { a = b; return 1; } return 0; }
template<class T> T SUM(const vector<T>& A) { return accumulate(ALL(A), T(0)); }
template<class T> vector<T> cumsum(const vector<T>& A, bool offset = true) { int N = A.size(); vector<T> S(N+1, 0); for (int i = 0; i < N; i++) { S[i+1] = S[i] + A[i]; } if (not offset) { S.erase(S.begin()); } return S; }
ll mod(ll x, ll m) { assert(m != 0); return (x % m + m) % m; }
ll ceil(ll a, ll b) { if (b < 0) { return ceil(-a, -b); } assert(b > 0); return (a < 0 ? a / b : (a + b - 1) / b); }
ll floor(ll a, ll b) { if (b < 0) { return floor(-a, -b); } assert(b > 0); return (a > 0 ? a / b : (a - b + 1) / b); }
ll powint(ll x, ll n) { assert(n >= 0); if (n == 0) { return 1; }; ll res = powint(x, n>>1); res *= res; if (n & 1) { res *= x; } return res; }
pair<ll,ll> divmod(ll a, ll b) { assert(b != 0); ll q = floor(a, b); return make_pair(q, a - q * b); }
ll bitlen(ll b) { if (b <= 0) { return 0; } return (64LL - __builtin_clzll(b)); }
ll digit_len(ll n) { assert(n >= 0); if (n == 0) { return 1; } ll sum = 0; while (n > 0) { sum++; n /= 10; } return sum; }
ll digit_sum(ll n) { assert(n >= 0); ll sum = 0; while (n > 0) { sum += n % 10; n /= 10; } return sum; }
ll digit_prod(ll n) { assert(n >= 0); if (n == 0) { return 0; } ll prod = 1; while (n > 0) { prod *= n % 10; n /= 10; } return prod; }
ll xor_sum(ll x) { assert(0 <= x); switch (x % 4) { case 0: return x; case 1: return 1; case 2: return x ^ 1; case 3: return 0; } assert(false); }
string toupper(const string& S) { string T(S); for (int i = 0; i < (int)T.size(); i++) { T[i] = toupper(T[i]); } return T; }
string tolower(const string& S) { string T(S); for (int i = 0; i < (int)T.size(); i++) { T[i] = tolower(T[i]); } return T; }
int a2i(const char& c) { assert(islower(c)); return (c - 'a'); }
int A2i(const char& c) { assert(isupper(c)); return (c - 'A'); }
int d2i(const char& d) { assert(isdigit(d)); return (d - '0'); }
char i2a(const int& i) { assert(0 <= i && i < 26); return ('a' + i); }
char i2A(const int& i) { assert(0 <= i && i < 26); return ('A' + i); }
char i2d(const int& i) { assert(0 <= i && i <= 9); return ('0' + i); }
using P = pair<ll,ll>;
using VP = vector<P>;
using VVP = vector<VP>;
using VS = vector<string>;
using VVS = vector<VS>;
using VI = vector<int>;
using VVI = vector<VI>;
using VVVI = vector<VVI>;
using VLL = vector<ll>;
using VVLL = vector<VLL>;
using VVVLL = vector<VVLL>;
using VB = vector<bool>;
using VVB = vector<VB>;
using VVVB = vector<VVB>;
using VD = vector<double>;
using VVD = vector<VD>;
using VVVD = vector<VVD>;
using VLD = vector<ld>;
using VVLD = vector<VLD>;
using VVVLD = vector<VVLD>;
const ld EPS = 1e-10;
const ld PI  = acosl(-1.0);
constexpr ll MOD = 1000000007;
// constexpr ll MOD = 998244353;
constexpr int inf = (1 << 30) - 1;   // 1073741824 - 1
constexpr ll INF = (1LL << 62) - 1;  // 4611686018427387904 - 1
// --------------------------------------------------------
// #include <atcoder/all>
// using namespace atcoder;


int main() {
    ios::sync_with_stdio(false);
    cin.tie(nullptr);
    cout << fixed << setprecision(15);

    ll H, W; input(W, H);
    VVLL A(H,VLL(W)); REP(h,H) REP(w,W) input(A[h][w]);

    auto on_grid = [&](int h, int w) -> bool { return (0 <= h && h < H && 0 <= w && w < W); };
    const VI dh4 = {0, 1, 0,-1};  // 右,下,左,上
    const VI dw4 = {1, 0,-1, 0};

    auto is_kadomatsu = [&](int a, int b, int c) -> bool {
        if (a == -1) return true;
        if (a == b || b == c || c == a) return false;
        if (a < b && b > c) return true;
        if (a > b && b < c) return true;
        return false;
    };

    map<tuple<int,int,int>,int> mp;  // (h, w, prev)
    queue<tuple<int,int,int>> q;
    mp[{0, 0, -1}] = 0; q.emplace(0, 0, -1);
    int ans = inf;
    while (not q.empty()) {
        auto [h, w, a] = q.front(); q.pop();
        if (h == H-1 && w == W-1) chmin(ans, mp[{h, w, a}]);
        int b = A[h][w];
        REP(d,4) {
            int hh = h + dh4[d];
            int ww = w + dw4[d];
            if (not on_grid(hh,ww)) continue;
            int c = A[hh][ww];
            if (not is_kadomatsu(a, b, c)) continue;
            if (not mp.count({hh, ww, b})) {
                mp[{hh, ww, b}] = mp[{h, w, a}] + 1;
                q.emplace(hh, ww, b);
            }
        }
    }

    if (ans == inf) ans = -1;
    print(ans);

    return 0;
}