for (int p=cur[i];p<v[i].size();p++) {
	cur[i] = p;
	if (v[i][p].u&&dis[v[i][p].i]==dis[i]+1) {
		ll flow = dfs(v[i][p].i, min(u, v[i][p].u));
		u-=flow, all+=flow;
		v[i][p].u-=flow;
		v[v[i][p].i][v[i][p].back].u+=flow;
		if (!u) return all;
	}
}

这个59ms

for (int p=cur[i];p<v[i].size();p++) {
	cur[i] = p+1;
	if (v[i][p].u&&dis[v[i][p].i]==dis[i]+1) {
		ll flow = dfs(v[i][p].i, min(u, v[i][p].u));
		u-=flow, all+=flow;
		v[i][p].u-=flow;
		v[v[i][p].i][v[i][p].back].u+=flow;
		if (!u) return all;
	}
}

这个659ms，仅仅改了这一个地方

cur[i] = p+1;

理应说来应该更快啊，为啥慢了好多？