FWT板子

板子：

#include<cmath>
#include<cstdio>
#include<cstring>
#include<algorithm>
using namespace std;
typedef long long ll;
const int N = ;
const int MOD = ;
template<typename T>
inline void read(T&x)
{
    T f = ,c = ;char ch=getchar();
    while(ch<''||ch>''){if(ch=='-')f=-;ch=getchar();}
    while(ch>=''&&ch<=''){c=c*+ch-'';ch=getchar();}
    x = f*c;
}
int n;
ll inv_2=(MOD+)/;
ll a[(<<N)+],b[(<<N)+],c[(<<N)+];
void Mod(ll&x){if(x>=MOD)x-=MOD;}
void fwt_or(ll *a,int len,int k)
{
    for(int i=;i<len;i<<=)
        for(int j=;j<len;j+=(i<<))
            for(int o=;o<i;o++)
            {
                if(k==)Mod(a[j+o+i]+=a[j+o]);
                else    Mod(a[j+o+i]+=-a[j+o]+MOD);
            }
}
void fwt_and(ll *a,int len,int k)
{
    for(int i=;i<len;i<<=)
        for(int j=;j<len;j+=(i<<))
            for(int o=;o<i;o++)
            {
                if(k==)Mod(a[j+o]+=a[j+o+i]);
                else    Mod(a[j+o]+=-a[j+o+i]+MOD);
            }
}
void fwt_xor(ll *a,int len,int k)
{
    for(int i=;i<len;i<<=)
        for(int j=;j<len;j+=(i<<))
            for(int o=;o<i;o++)
            {
                int w1 = a[j+o],w2 = a[j+o+i];
                a[j+o] = (w1+w2)%MOD;
                a[j+o+i] = (w1-w2+MOD)%MOD;
                if(k==-)a[j+o]=a[j+o]*inv_2%MOD,a[j+o+i]=a[j+o+i]*inv_2%MOD;
            }
}
void mul()
{
    for(int i=;i<(<<n);i++)
        c[i] = a[i]*b[i]%MOD;
}
void print()
{
    for(int i=;i<(<<n);i++)
        printf("%lld ",c[i]);
    puts("");
}
int main()
{
    read(n);int len = (<<n);
    for(int i=;i<len;i++)
        read(a[i]);
    for(int i=;i<len;i++)
        read(b[i]);
    fwt_or(a,len,),fwt_or(b,len,);
    mul();
    fwt_or(c,len,-);
    print();
    fwt_or(a,len,-),fwt_or(b,len,-);
    fwt_and(a,len,),fwt_and(b,len,);
    mul();
    fwt_and(c,len,-);
    print();
    fwt_and(a,len,-),fwt_and(b,len,-);
    fwt_xor(a,len,),fwt_xor(b,len,);
    mul();
    fwt_xor(c,len,-);
    print();
    fwt_xor(a,len,-),fwt_xor(b,len,-);
    return ;
}