#pragma GCC optimize("Ofast,unroll-loops")
#pragma GCC target("avx")
#include <bits/stdc++.h>
using namespace std;

#include <vector>
#include <iostream>
#include <bitset>

#include <vector>

struct residual_graph {
  using cap_type = long long;
  struct edge {
    int to;
    cap_type cap;
    int rev;
    edge(int t = 0, cap_type c = 0, int r = 0): to(t), cap(c), rev(r) {}
  };
  int N;
  std::vector<int> iter;
  std::vector<std::tuple<int, int, cap_type>> pool;
  std::vector<edge> edges;
  residual_graph(int N): N(N), iter(N + 1) {}

  void add_edge(int from, int to, cap_type cap, cap_type rev_cap) {
    iter[from]++;
    iter[to]++;
    pool.emplace_back(from, to, cap);
  }
  void build() {
    for(int i = 0; i < N; i++) {
      iter[i + 1] += iter[i];
    }
    edges.resize(pool.size() * 2);
    for(auto&& p: pool) {
      int fi = --iter[std::get<0>(p)];
      int ti = --iter[std::get<1>(p)];
      edges[fi] = edge(std::get<1>(p), std::get<2>(p), ti);
      edges[ti] = edge(std::get<0>(p), 0, fi);
    }
  }
  inline edge& operator[](int i) { return edges[i]; }
};

struct goldberg_tarjan {
  using cap_type = long long;
  struct edge {
    int to;
    cap_type cap;
    edge(int t, cap_type c, int r): to(t), cap(c) {}
  };

  int N;
  residual_graph G;
  std::vector<cap_type> exc;
  std::vector<int> h;
  std::vector<std::vector<int>> que;
  std::vector<int> qi;
  std::vector<int> ei;
  std::bitset<2048> inque;
  const cap_type INF = 1e18;

  goldberg_tarjan(int n): N(n), G(n) {}
  
  void add_edge(int from, int to, cap_type cap, cap_type rev_cap = 0) {
    G.add_edge(from, to, cap, rev_cap);
  }
  void build() {
    G.build();
  }

  void push(int from, int ei) {
    cap_type fl = std::min(exc[from], G[ei].cap);
    G[ei].cap -= fl;
    G[G[ei].rev].cap += fl;
    exc[from] -= fl;
    exc[G[ei].to] += fl;
  }

  void relabel(int v) {
    h[v] = N;
    for(int i = G.iter[v]; i < G.iter[v + 1]; i++) {
      auto& e = G[i];
      if(e.cap > 0 && h[v] > h[e.to] + 1) {
        ei[v] = i;
        h[v] = h[e.to] + 1;
      }
    }
  }

  int global_relabeling(int t) {
    std::fill(std::begin(h), std::end(h), N);
    for(int i = 0;i < N;i++) {
      que[i].clear();
    }
    inque.reset();
    int i = 0, qr = 0;
    std::vector<int>& Q = qi;
    Q[qr++] = t;
    h[t] = 0;
    int hi = 0;
    while(i < qr) {
      int v = Q[i++];
      hi = h[v];
      if(exc[v] > 0 && v != t) {
        que[h[v]].emplace_back(v);
        inque.set(v);
      }
      for(int gi = G.iter[v]; gi < G.iter[v + 1]; gi++) {
        auto& e = G[gi];
        if(G[e.rev].cap > 0 && h[v] + 1 < h[e.to]) {
          h[e.to] = h[v] + 1;
          Q[qr++] = e.to;
        }
      }
    }
    std::copy(std::begin(G.iter), std::begin(G.iter) + N, std::begin(ei));
    std::fill(std::begin(qi), std::end(qi), 0);
    return hi;
  }

  void gap_relabeling(int g) {
    for(int v = 0; v < N; v++) {
      if(h[v] > g) {
        h[v] = N;
      }
    }
  }


  cap_type max_flow(int s, int t) {
    exc.assign(N, 0);
    exc[s] = INF;
    h.assign(N, 0);
    int cnt = 0;

    que.resize(N);
    qi.assign(N, 0);
    ei.resize(N);

    global_relabeling(t);

    if(h[s] == N) return 0;

    for(int di = h[s]; di >= 0;) {
      if(qi[di] == que[di].size()) { di--; continue; }
      int v = que[di][qi[di]++];
      inque.reset(v);
      if(exc[v] == 0 || v == t) continue;
      for(int& i = ei[v]; i < G.iter[v + 1]; i++) {
        auto& e = G[i];
        if(e.cap > 0 && h[v] == h[e.to] + 1) {
          push(v, i);
          if(exc[e.to] > 0 && e.to != t && !inque.test(e.to)) {
            que[h[e.to]].emplace_back(e.to);
            inque.set(e.to);
          }
          if(exc[v] == 0) break;
        }
      }
      if(exc[v] == 0) continue;
      relabel(v);
      if(h[v] < N) {
        di = h[v];
        que[h[v]].emplace_back(v);
        inque.set(v);
      }

      if(qi[di] == que[di].size()) {
        gap_relabeling(di);
        di--;
      }

      if(++cnt % (N * 4) == 0) {
        di = global_relabeling(t);
      }
    }
    return exc[t];
  }
};
#include <bits/stdc++.h>
using namespace std;
using i64 = long long;
#define rep(i,s,e) for(i64 (i) = (s);(i) < (e);(i)++)
#define all(x) x.begin(),x.end()

template<class T>
static inline std::vector<T> ndvec(size_t&& n, T val) noexcept {
  return std::vector<T>(n, std::forward<T>(val));
}

template<class... Tail>
static inline auto ndvec(size_t&& n, Tail&&... tail) noexcept {
  return std::vector<decltype(ndvec(std::forward<Tail>(tail)...))>(n, ndvec(std::forward<Tail>(tail)...));
}

template<class T, class Cond>
struct chain {
  Cond cond; chain(Cond cond) : cond(cond) {}
  bool operator()(T& a, const T& b) const {
    if(cond(a, b)) { a = b; return true; }
    return false;
  }
};
template<class T, class Cond>
chain<T, Cond> make_chain(Cond cond) { return chain<T, Cond>(cond); }

int main() {
  cin.tie(nullptr);
  std::ios::sync_with_stdio(false);
  int H, W;
  cin >> H >> W;

  goldberg_tarjan gt(H + W + 2);
  int s = H + W;
  int t = s + 1;
  vector<i64> A(H, 0);
  rep(i,0,H) rep(j,0,W) {
    i64 g;
    cin >> g;
    A[i] += g;
    gt.add_edge(i, H + j, g);
  }
  i64 sum = 0;
  rep(i,0,H) {
    i64 r;
    cin >> r;
    i64 MIN = std::min(A[i], r);
    sum += r - MIN;
    gt.add_edge(s, i, A[i] - MIN);
  }
  rep(i,0,W) {
    i64 r;
    cin >> r;
    sum += r;
    gt.add_edge(H + i, t, r);
  }
  cout << sum - gt.max_flow(s, t) << "\n";
}