MLE or TLE on 4，但不知道怎么优化。

#pragma GCC optimize(2)
#include <bits/stdc++.h>
#include <ext/pb_ds/assoc_container.hpp>
#include <ext/pb_ds/hash_policy.hpp>

using namespace std;
using namespace __gnu_pbds;

inline int read()
{
    int s = 0, w = 1;
    char c = getchar();
    for ( ; !isdigit(c); c = getchar())  if (c == '-') w = -1;
    for ( ; isdigit(c); c = getchar())   s = (s << 1) + (s << 3) + (c ^ 48);
    return s * w;
}

const int N = 1e6 + 10;

bool Mbe;
int n, q;
struct Node
{
	int l, r;
} tr[N * 21]; 
gp_hash_table<int, bool> T;
int w[N];
int root[N], idx;
int back[N];
int lop[N], lv[N], top, i;
bool Med;

int build(int l, int r)
{
	int p = ++ idx;
	if (l == r) return p;
	int mid = l + r >> 1;
	tr[p].l = build(l, mid), tr[p].r = build(mid + 1, r);
	return p;
}

int add(int p, int l, int r, int x, int v)
{
	int q = ++ idx; tr[q] = tr[p];
	if (l == r)
	{
		T[q] = T[p];
		if (v == 1 && T[q] == 0) w[i] ++ ;
		if (v == -1 && T[q] == 1) w[i] -- ;
		T[q] += v; return q;
	}
	int mid = l + r >> 1;
	if (x <= mid) tr[q].l = add(tr[p].l, l, mid, x, v);
	else tr[q].r = add(tr[p].r, mid + 1, r, x, v);
	return q;
}

int main()
{
//	freopen(".in", "r", stdin);
//	freopen(".out", "w", stdout);
	
	fprintf(stderr, "%.3lf MB\n", (&Mbe - &Med) / 1048576.0);
	root[0] = build(1, N - 1);
	scanf("%d", &q); int len;
	for (i = 1, len = 0; i <= q; i ++ )
	{
		char op[2]; scanf("%s", op);
		if (*op == '+')
		{
			int x; scanf("%d", &x);
			w[i] = w[i - 1]; root[i] = add(root[i - 1], 1, N - 1, x, 1); len ++ ;
			back[len] = i; lop[ ++ top] = 1; lv[top] = x;
		}
		else if (*op == '-')
		{
			int k; scanf("%d", &k);
			len -= k; root[i] = root[back[len]]; w[i] = w[back[len]];
			back[len] = i; lop[ ++ top] = 2; lv[top] = k;
		}
		else if (*op == '!')
		{
			if (lop[top] == 1)
			{
				w[i] = w[i - 1]; root[i] = add(root[i - 1], 1, N - 1, lv[top], -1);
				len -- ; back[len] = i;
			}
			else 
			{
				len += lv[top]; root[i] = root[back[len]]; w[i] = w[back[len]];
				back[len] = i; 
			}
			top -- ;
		}
		else 
		{
			root[i] = root[i - 1]; w[i] = w[i - 1]; 
			printf("%d\n", w[i]);
		}
	}	
	
	return 0;
}