这是代码，本人已经优化到极限了

#include <cstdio>
#include <queue>
#include <cstring>
#include <algorithm>
using namespace std;

struct num_with_xy
{
	int x, y;
	int num;
}t[10010];
struct node
{
	int x, y;
};
int n, m;
char mp[1010][1010];
int vis[1010][1010], walk[4][2] = {{0, 1}, {1, 0}, {0, -1}, {-1, 0}};
queue<node> q;
void bfs(int x, int y)
{
	memset(vis, -1, sizeof vis);
	q.push({x, y});
	vis[x][y] = 0;
	while(!q.empty())
	{
		int xx = q.front().x, yy = q.front().y;
		q.pop();
		for(int i = 0; i < 4; i++)
		{
			int u = xx + walk[i][0], v = yy + walk[i][1];
			if(1 <= u && u <= n && 1 <= v && v <= m && vis[u][v] == -1 && mp[u][v] != '1')
			{
				vis[u][v] = vis[xx][yy] + 1;
				q.push({u, v});
			}
		}
	}
}

int main()
{
	scanf("%d %d", &m, &n);
	for(int i = 1; i <= n; i++)
		for(int j = 1; j <= m; j++) scanf("%s", &mp[i][j]);
	int minn = 1e9, cur = 0;
	int tx = 0, ty = 0, ex = 0, ey = 0;
	for(int i = 1; i <= n; i++)
	{
		for(int j = 1; j <= m; j++)
		{
			if(mp[i][j] == '2') bfs(i, j);
			if(mp[i][j] == '3') ex = i, ey = j;
		}
	}
	for(int i = 1; i <= n; i++)
	{
		for(int j = 1; j <= m; j++)
			if(mp[i][j] == '4' && vis[i][j] != -1)
			{
				t[++cur].num = vis[i][j];
				t[cur].x = i;
				t[cur].y = j;
			}
	}
	for(int i = 1; i <= cur; i++)
	{
		bfs(t[i].x, t[i].y);
		minn = min(minn, t[i].num + vis[ex][ey]);
	}
	printf("%d", minn);
	return 0;
}