牛客练习赛50 F tokitsukaze and Another Protoss and Zerg（分治NTT）

一些废话：
写完上一题（分治FFT）之后记起之前牛客写的这道题，题解是用启发式NTT，每次选两个项数最小的合并，当时没有整理NTT的模板所以就先mark了，今天受到分治FFT的启发，想用分治NTT来解决这道题。
复杂度分析：一共会分出log(n)层，每层的项数总和数量级是$\Sigma a_i$,ntt复杂度nlogn，所以总复杂度是$O(nlo{g}^{2}n)$.
这题也是用生成函数，每一场的生成函数为：$(2^{b_i}-1)+{\sum }_{i=1}^{a_i}{C}_{a_i}^i{x}^i$
把每一场的函数乘起来之后$x^i$的系数就是恰好取i个的答案。
（一定要好好审题

#include<bits/stdc++.h>
#define ll long long
using namespace std;
const int maxn = 2e5 + 50;
const ll mod = 998244353;
ll qm(ll a, ll b){ll res = 1; while(b){if(b&1) res = res*a%mod; a = a*a%mod; b >>= 1; } return res; }
ll mulwn[maxn<<2], invwn[maxn<<2];
void INIT(){
    for(ll i=1;i < maxn*4;i<<=1) mulwn[i]=qm(3,(mod-1)/i);
    for(ll i=1;i < maxn*4;i<<=1) invwn[i]=qm(mulwn[i],mod-2);
}
struct FFT {
    ll n, m, rev[maxn << 2];
    ll a[maxn << 2], b[maxn << 2];
    void init(int len) {
        for (n = 1, m = 0; n < len + len; n <<= 1, m++);
        for (int i = 0; i < n; ++i) {
            rev[i] = (rev[i >> 1] >> 1) | (i & 1) << (m - 1);
            a[i] = 0;
            b[i] = 0;
        }
    }
    void ntt(ll *a, int f) {//
        for (int i = 0; i < n; ++i)if (i < rev[i])swap(a[i], a[rev[i]]);

        for (int k = 2; k <= n; k <<= 1) {
            ll wn=(f>0)?mulwn[k]:invwn[k];
            int mid = k>>1;
            for(int i = 0; i < n; i += k){
                ll w = 1;
                for(int j = 0; j < mid; ++j, w = w*wn%mod){
                    ll temp = (w*a[i+j+mid])%mod;
                    a[i+j+mid] = (a[i+j]-temp+mod)%mod;
                    a[i+j] = (a[i+j]+temp)%mod;
                }
            }
        }
        return;
    }
    void Calculate() {
        ntt(a, 1); ntt(b, 1);
        for (int i = 0; i < n; ++i)a[i] = a[i]*b[i]%mod;//记得取模
        ntt(a, -1);
        ll invl = qm(n, mod-2);
        for(int i = 0; i < n; ++i){
            a[i] = a[i]*invl%mod;
        }
    }
} F;
int n;
ll p[maxn], inv[maxn];
ll a[maxn], b[maxn];
ll p2[maxn];
void init()
{
    cin>>n;
    for(int i = 1; i <= n; ++i) scanf("%lld", &a[i]);
    for(int i = 1; i <= n; ++i) scanf("%lld", &b[i]);
}
ll Com(int n, int m){
    ll ans = p[n]*inv[m]%mod*inv[n-m]%mod;
    return ans;
}
vector<ll> w[maxn<<2];
void sol(int rt, int l, int r){
    w[rt].clear();
    if(l == r){
        w[rt].push_back( (p2[b[l]]-1+mod)%mod );
        for(int i = 1; i <= a[l]; ++i){
            w[rt].push_back(Com(a[l], i));
        }
        return;
    }
    int mid = (l+r)>>1;
    sol(rt<<1, l, mid); sol(rt<<1|1, mid+1, r);
    int len = w[rt<<1].size() + w[rt<<1|1].size()-1;
    F.init(len);
    for(int i = 0; i < w[rt<<1].size(); ++i) F.a[i] = w[rt<<1][i];
    for(int i = 0; i < w[rt<<1|1].size(); ++i) F.b[i] = w[rt<<1|1][i];
    F.Calculate();
    for(int i = 0; i < len; ++i){
        w[rt].push_back(F.a[i]%mod);
    }
}
int main(){
    p2[0] = p[0] = inv[0] = 1;
    for(int i = 1; i < maxn; ++i) p[i] = p[i-1]*i%mod, inv[i] = qm(p[i], mod-2), p2[i] = p2[i-1]*2%mod;
    INIT();
    init();
    sol(1, 1, n);
    for(int i = 0; i < w[1].size(); ++i) {
        if(i > 0) printf(" ");
        printf("%lld", (w[1][i]+mod)%mod);
    }printf("\n");
}