[Ynoi2016]这是我自己的发明 dfs序+树上倍增+莫队

Description
给一个树，n个点，有点权，初始根是1。
m 个操作，每次操作：
1.将树根换为 x。
2.给出两个点 x，y，从 x 的子树中选每一个点，y 的子树中选每一个点，如果两个点点权相等，ans++，求 ans。

---

Sample Input
5 5
1 2 3 4 5
1 2
1 3
3 4
3 5
2 4 5
2 1 5
2 3 5
1 5
2 4 5

---

Sample Output
0
1
1
1

---

这题好码农啊。。。
这题一看，换根操作，马上就想到遥远的国度，于是你就dfs序一下，然后你就想到一个简单的询问那道题，然后就莫队做。。。
一开始我把它拆成16个操作，结果被一条链那个数据点卡了。。。
后来肉老师一波小提示，结果发现可以拆成9个操作，然后就A了，但时间复杂度怎么还是这么高啊，全OJ垫底？？？

---

#include <cmath>
#include <cstdio>
#include <cstring>
#include <algorithm>

using namespace std;
int read() {
    int x = 0, f = 1; char ch = getchar();
    while(ch < '0' || ch > '9') {if(ch == '-') f = -1; ch = getchar();}
    while(ch >= '0' && ch <= '9') {x = x * 10 + ch - '0'; ch = getchar();}
    return x * f;
}
int _min(int x, int y) {return x < y ? x : y;}

struct edge {
    int x, y, next;
} e[210000]; int len, last[110000];
struct node {
    int l, r, kk, opt, ll;
} q[8100000];
int n, id, root, ll[110000], rr[110000], dep[110000], fa[110000][20];
int ans, a[110000], pos[110000], b[110000];
int p1[110000], p2[110000];
int gg[510000];
bool v[510000];

bool cmp(node a, node b) {
    if(a.kk == b.kk) return a.r < b.r;
    return a.kk < b.kk;
}

void ins(int x, int y) {
    e[++len].x = x; e[len].y = y;
    e[len].next = last[x]; last[x] = len;
}

int erfen(int x) {
    int l = 1, r = n, ans;
    while(l <= r) {
        int mid = (l + r) / 2;
        if(x <= b[mid]) r = mid - 1, ans = mid;
        else l = mid + 1;
    }
    return ans;
}

void dfs(int x) {
    ll[x] = ++id;
    pos[id] = a[x];
    for(int i = 1; i <= 18; i++) fa[x][i] = fa[fa[x][i - 1]][i - 1];
    for(int k = last[x]; k; k = e[k].next) {
        int y = e[k].y;
        if(y != fa[x][0]) {
            fa[y][0] = x;
            dep[y] = dep[x] + 1;
            dfs(y);
        }
    }
    rr[x] = id;
}

int jump(int x, int h) {
    for(int i = 18; i >= 0; i--)
        if((1 << i) <= h)
            x = fa[x][i], h -= (1 << i);
    return x;
}

void hl(int k, int o) {p1[pos[k]] += o; ans += p2[pos[k]] * o;}
void hr(int k, int o) {p2[pos[k]] += o; ans += p1[pos[k]] * o;}

int main() {
    int m; n = read(); m = read();
    for(int i = 1; i <= n; i++) a[i] = read(), b[i] = a[i];
    sort(b + 1, b + n + 1);
    for(int i = 1; i <= n; i++) a[i] = erfen(a[i]);
    for(int i = 1; i < n; i++) {
        int x = read(), y = read();
        ins(x, y); ins(y, x);
    }
    dfs(1); root = 1;
    int len = 0;
    for(int i = 1; i <= m; i++) {
        int opt = read();
        if(opt == 1) root = read();
        else {
            v[i] = 1;
            int x = read(), y = read();
            if(ll[root] > ll[x] && ll[root] <= rr[x] && ll[root] > ll[y] && ll[root] <= rr[y]) {
                int o1 = jump(root, dep[root] - dep[x] - 1), o2 = jump(root, dep[root] - dep[y] - 1);
                int l1 = ll[o1] - 1, r1 = rr[o1];
                int l2 = ll[o2] - 1, r2 = rr[o2];
                q[++len].l = n; q[len].r = n; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
                q[++len].l = r2; q[len].r = n; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
                q[++len].l = l2; q[len].r = n; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
                q[++len].l = r1; q[len].r = n; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
                q[++len].l = r1; q[len].r = r2; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
                q[++len].l = l2; q[len].r = r1; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
                q[++len].l = l1; q[len].r = n; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
                q[++len].l = l1; q[len].r = r2; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
                q[++len].l = l1; q[len].r = l2; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
            }
            else if(ll[root] > ll[x] && ll[root] <= rr[x]) {
                int o1 = jump(root, dep[root] - dep[x] - 1);
                int l1 = ll[o1] - 1, r1 = rr[o1];
                int l2 = ll[y] - 1, r2 = rr[y];
                if(y == root) l2 = 0, r2 = n;
                q[++len].l = r2; q[len].r = n; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
                q[++len].l = r1; q[len].r = r2; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
                q[++len].l = l1; q[len].r = r2; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
                q[++len].l = l2; q[len].r = n; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
                q[++len].l = l2; q[len].r = r1; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
                q[++len].l = l1; q[len].r = l2; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
            }
            else if(ll[root] > ll[y] && ll[root] <= rr[y]) {
                int o2 = jump(root, dep[root] - dep[y] - 1);
                int l1 = ll[x] - 1, r1 = rr[x];
                if(x == root) l1 = 0, r1 = n;
                int l2 = ll[o2] - 1, r2 = rr[o2];
                q[++len].l = r1; q[len].r = n; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
                q[++len].l = l1; q[len].r = n; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
                q[++len].l = r1; q[len].r = r2; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
                q[++len].l = l1; q[len].r = r2; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
                q[++len].l = l2; q[len].r = r1; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
                q[++len].l = l1; q[len].r = l2; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
            }
            else {
                int l1 = ll[x] - 1, r1 = rr[x];
                if(x == root) l1 = 0, r1 = n;
                int l3 = ll[y] - 1, r3 = rr[y];
                if(y == root) l3 = 0, r3 = n;
                q[++len].l = r1; q[len].r = r3; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
                q[++len].l = r1; q[len].r = l3; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
                q[++len].l = l1; q[len].r = r3; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = -1;
                q[++len].l = l1; q[len].r = l3; if(q[len].l > q[len].r) swap(q[len].l, q[len].r);
                q[len].opt = i; q[len].ll = 1;
            }
        }
    }
    int pp = sqrt(n);
    for(int i = 1; i <= len; i++) q[i].kk = _min(q[i].l / pp + 1, pp + 1);
    sort(q + 1, q + len + 1, cmp);
    int nowl = 0, nowr = 0;
    for(int i = 1; i <= len; i++) {
        for(int j = nowr + 1; j <= q[i].r; j++) hr(j, 1);
        for(int j = nowr; j > q[i].r; j--) hr(j, -1);
        for(int j = nowl + 1; j <= q[i].l; j++) hl(j, 1);
        for(int j = nowl; j > q[i].l; j--) hl(j, -1);
        nowl = q[i].l, nowr = q[i].r;
        gg[q[i].opt] += ans * q[i].ll;
    }
    for(int i = 1; i <= m; i++) if(v[i]) printf("%d\n", gg[i]);
    return 0;
}