T1 TLE 求助

#include<bits/stdc++.h>
#define int long long
#define mod 998244353
using namespace std;
int read(){
    int sum=0,fish=1;
    char c=getchar_unlocked();
    while((c<'0'||c>'9')&&c!='-')c=getchar_unlocked();
    if(c=='-')fish=-1,c=getchar_unlocked();
    while(c>='0'&&c<='9')sum=sum*10+(c-'0'),c=getchar_unlocked();
    return sum*fish;
}
void print(int x){
    if(x<0)putchar_unlocked('-'),x=-x;
    if(x<10)putchar_unlocked(x+'0');
    else print(x/10),putchar_unlocked(x%10+'0');
}
int a[8005];
int A[8005],inv[8005];
int C(int a,int b){
    if(a<b||b<0)return 0;
    return A[a]*inv[b]%mod*inv[a-b]%mod;
}
int dp[2][8005];
void solve(){
    memset(dp,0,sizeof dp);
    int n=read(),k=read();
    for(int i=1;i<=n;i++)
    a[i]=-read();
    sort(a+1,a+1+n);
    vector<int>ve;
    for(int i=1;i<=n;i++)
    if(a[i]==a[i-1])ve[ve.size()-1]++;
    else ve.push_back(1);
    int sum=0,top=1;
    dp[0][0]=1;
    for(auto i:ve){
        dp[top][0]=0;
        for(int j=1;j<=k;j++)
        dp[top][j]=(dp[top^1][j]*C(sum+i-1,i)%mod+
                    dp[top^1][j-1]*C(sum+i-1,i-1)%mod)%mod*A[i]%mod;
        top^=1;
        sum+=i;
    }
    print(dp[top^1][k]);puts("");
}
int qpow(int a,int b=mod-2){
    int ans=1;
    while(b){
        if(b&1)ans=ans*a%mod;
        b>>=1;
        a=a*a%mod;
    }
    return ans;
}
signed main(){
    ios::sync_with_stdio(0);
    cin.tie(0);
    cout.tie(0);
    A[0]=1;
    for(int i=1;i<=8000;i++)
    A[i]=A[i-1]*i%mod;
    inv[8000]=qpow(A[8000])%mod;
    for(int i=8000;i;i--)
    inv[i-1]=inv[i]*i%mod;
    int t=read();
    while(t--)solve();
    return 0;
}