ソースコード

// #pragma GCC optimize ("Ofast") // #pragma GCC optimize ("unroll-loops") // #pragma GCC target ("avx,avx2,fma") #include <bits/stdc++.h> #define rep(i, a, b) for(int i = (a); i <= (b); i ++) using std::cin, std::cout, std::cerr; using ll = long long; const ll P = 998244353, G = 3; ll qpow(ll a, ll b) { ll r = 1; for(; b; b /= 2) { if(b & 1) r = r * a % P; a = a * a % P; } return r; } ll INV(ll x) { return qpow(x, P - 2); } using vl = std::vector<ll>; void ntt(vl &a, int op = 1) { // P: 模数, G: 原根	int n = a.size();	for (int i = 1, j = 0; i < n - 1; ++i) {	for (int s = n; j ^= s >>= 1, ~j & s;) {}	if (i < j) std::swap(a[i], a[j]);	}	for (int i = 1; i < n; i *= 2) {	ll u = qpow(G, (P - 1) / (i * 2));	if (op == -1) u = INV(u);	for (int j = 0; j < n; j += i * 2) {	ll w = 1;	for (int k = 0; k < i; ++k, w = w * u % P) {	int x = a[j + k], y = w * a[j + k + i] % P;	#define modto(x) ({if ((x) >= P) (x) -= P;})	a[j + k] = x + y; modto(a[j + k]);	a[j + k + i] = x - y + P; modto(a[j + k + i]);	#undef modto	}}}	if (op == -1) { ll inv = INV(n);	for (int i = 0; i < n; ++i) a[i] = a[i] * inv % P; } } vl multi(vl a, vl b) {	int m = a.size() + b.size() - 1;	int n = 1; while (n < m) n *= 2;	vl fa(n), fb(n), fc(n);	std::copy(a.begin(), a.end(), fa.begin()); ntt(fa);	std::copy(b.begin(), b.end(), fb.begin()); ntt(fb);	for (int i = 0; i < n; ++i) fc[i] = fa[i] * fb[i] % P;	ntt(fc, -1); fc.resize(m);	return fc; } int main() { std::ios::sync_with_stdio(false); int n, q; cin >> n >> q; std::vector<ll> a(n), b(n); rep(i, 0, n - 1) cin >> a[i]; while(q --) { int x; cin >> x; b[(n - x) % n] ++; } auto c = multi(a, b); for(int i = n; i < c.size(); i ++) c[i - n] += c[i]; rep(i, 0, n - 1) cout << c[i] << ' '; cout << '\n'; }