#947213 (Rust) No.2116 Making Forest Hard

• 問題: No.2116 Making Forest Hard
• ユーザー: akakimidori
• 提出日時: 2024-01-29 23:20:27
• 言語: Rust (1.83.0 + proconio)
• 結果: CE (最新)
  AC
  (最初)
• 実行時間: -
• コード長: 27,285 bytes
• コンパイル時間: 30,358 ms
• コンパイル使用メモリ: 6,948 KB
• 最終ジャッジ日時: 2024-12-30 22:38:27
• 合計ジャッジ時間: 34,492 ms
• ジャッジサーバーID （参考情報）: judge5 / judge6

コンパイルメッセージ

コンパイルが30秒の制限時間を超えました

ソースコード

// dp[v][a]: 部分木vでvを含みそれが最上位な連結成分について
// 連結成分のmaxがa であるようなものについての
// ... 何を管理するとdpできる？
// way: そうなるような切り方の個数
// sum: そうなる切り方についてのvを含む連結成分サイズの和
// というのを管理できてるとして
// 根vに親pが生成される時
// dp[p][A_p].0 = dp[v][*].0 + dp[v][A <= A_p].0
// dp[p][A_p].1 = 2 * dp[v][*].0 + dp[v][A <= A_p].1
//
// dp[p][x] = (0, 0) (x < A_p)
// dp[p][x].0 = dp[v][x].0
// dp[p][x].1 = dp[v][x].0 + dp[v][x].1 (x > A_p)
//
// 子にuが生えた時

type M = ModInt<998244353>;

fn run() {
    input! {
        n: usize,
        a: [u32; n],
        e: [(usize1, usize1); n - 1],
    }
    let a = a
        .into_iter()
        .enumerate()
        .map(|p| (p.1, p.0))
        .collect::<Vec<_>>();
    let mut g = vec![vec![]; n];
    for &(a, b) in e.iter() {
        g[a].push(b);
        g[b].push(a);
    }
    let root = 0;
    let mut topo = vec![root];
    let mut parent = vec![n; n];
    for i in 0..n {
        let v = topo[i];
        for u in g[v].clone() {
            g[u].retain(|p| *p != v);
            topo.push(u);
            parent[u] = v;
        }
    }
    let mut size = vec![1usize; n];
    for &v in topo.iter().rev() {
        g[v].sort_by_key(|u| !size[*u]);
        for &u in g[v].iter() {
            size[v] += size[u];
        }
    }
    let mut memo: Vec<Vec<((u32, usize), Dual<M>)>> = vec![vec![]; n];
    let mut ans = M::zero();
    for &v in topo.iter().rev() {
        if v != root && g[parent[v]][0] == v {
            continue;
        }
        let mut z = vec![];
        let mut pos = v;
        let mut path = vec![];
        loop {
            z.push(a[pos]);
            path.push(pos);
            for &u in g[pos].iter().skip(1) {
                z.extend(memo[u].iter().map(|p| p.0));
            }
            if g[pos].is_empty() {
                break;
            }
            pos = g[pos][0];
        }
        z.sort();
        let mut seg = LazySegmentTree::new(z.len(), R);
        for &v in path.iter().rev() {
            let x = z.binary_search(&a[v]).unwrap();
            let Dual(w, s) = seg.find(0, x).0;
            let Dual(all, _) = seg.find(0, z.len()).0;
            seg.update(0, x, Dual::zero());
            seg.update(x + 1, z.len(), Dual::new(M::one(), M::one()));
            let mut p = Dual::new(all + w, all + s + w);
            if v == *path.last().unwrap() {
                p = Dual::new(M::one(), M::one());
            }
            seg.set_at(x, (p, p * Dual::new(M::new(a[v].0), M::zero())));
            for &u in g[v].iter().skip(1) {
                let c = std::mem::take(&mut memo[u]);
                let mut memo = vec![Dual::zero(); c.len()];
                let mut pos = vec![0; c.len()];
                for (i, &(key, _)) in c.iter().enumerate() {
                    let x = z.binary_search(&key).unwrap();
                    memo[i] = seg.find(0, x).0;
                    pos[i] = x;
                }
                pos.insert(0, 0);
                pos.push(z.len());
                let way = c.iter().fold(M::zero(), |s, &(_, a)| s + a.0);
                let mut pre = Dual::new(way, M::zero());
                for (i, x) in pos.windows(2).enumerate() {
                    let (l, r) = (x[0], x[1]);
                    seg.update(l, r, pre);
                    if i < c.len() {
                        let c = c[i].1;
                        pre = pre + c;
                    }
                }
                for i in 0..c.len() {
                    let a = M::new(c[i].0 .0);
                    let c = c[i].1;
                    let memo = memo[i];
                    let pos = pos[i + 1];
                    seg.set_at(pos, (c * memo, c * memo * Dual::new(a, M::zero())));
                }
            }
            if v == root {
                ans += seg.find(0, z.len()).1 .1;
            } else {
                ans += seg.find(0, z.len()).1 .1 * M::new(2).pow((n - 1 - size[v]) as u64);
            }
        }
        memo[v] = (0..z.len()).map(|x| (z[x], seg.find(x, x + 1).0)).collect();
    }
    println!("{}", ans);
}

struct R;
impl TE for R {
    type T = (Dual<M>, Dual<M>);
    type E = Dual<M>;
    fn fold(&self, l: &Self::T, r: &Self::T) -> Self::T {
        (l.0 + r.0, l.1 + r.1)
    }
    fn eval(&self, x: &Self::T, f: &Self::E) -> Self::T {
        (x.0 * *f, x.1 * *f)
    }
    fn merge(&self, g: &Self::E, h: &Self::E) -> Self::E {
        *g * *h
    }
    fn e(&self) -> Self::T {
        (Dual::zero(), Dual::zero())
    }
    fn id(&self) -> Self::E {
        Dual::one()
    }
}

fn main() {
    run();
}

#[derive(Clone, Copy, Default, Debug)]
pub struct Dual<T>(T, T);

impl<T> Dual<T> {
    pub fn new(a: T, b: T) -> Self {
        Self(a, b)
    }
}

impl<T> Zero for Dual<T>
where
    T: Zero,
{
    fn zero() -> Self {
        Self::new(T::zero(), T::zero())
    }
    fn is_zero(&self) -> bool {
        self.0.is_zero() && self.1.is_zero()
    }
}

impl<T> One for Dual<T>
where
    T: One + Zero + Clone,
{
    fn one() -> Self {
        Self::new(T::one(), T::zero())
    }
    fn is_one(&self) -> bool {
        self.0.is_one() && self.1.is_zero()
    }
}

impl<T> Add for Dual<T>
where
    T: Add<Output = T>,
{
    type Output = Self;
    fn add(self, rhs: Self) -> Self {
        Self::new(self.0 + rhs.0, self.1 + rhs.1)
    }
}

impl<T> Sub for Dual<T>
where
    T: Sub<Output = T>,
{
    type Output = Self;
    fn sub(self, rhs: Self) -> Self {
        Self::new(self.0 - rhs.0, self.1 - rhs.1)
    }
}

impl<T> Mul for Dual<T>
where
    T: Clone + Add<Output = T> + Mul<Output = T>,
{
    type Output = Self;
    fn mul(self, rhs: Self) -> Self {
        Self::new(
            self.0.clone() * rhs.0.clone(),
            self.0 * rhs.1 + self.1 * rhs.0,
        )
    }
}

// ---------- begin input macro ----------
// reference: https://qiita.com/tanakh/items/0ba42c7ca36cd29d0ac8
#[macro_export]
macro_rules! input {
    (source = $s:expr, $($r:tt)*) => {
        let mut iter = $s.split_whitespace();
        input_inner!{iter, $($r)*}
    };
    ($($r:tt)*) => {
        let s = {
            use std::io::Read;
            let mut s = String::new();
            std::io::stdin().read_to_string(&mut s).unwrap();
            s
        };
        let mut iter = s.split_whitespace();
        input_inner!{iter, $($r)*}
    };
}

#[macro_export]
macro_rules! input_inner {
    ($iter:expr) => {};
    ($iter:expr, ) => {};
    ($iter:expr, $var:ident : $t:tt $($r:tt)*) => {
        let $var = read_value!($iter, $t);
        input_inner!{$iter $($r)*}
    };
}

#[macro_export]
macro_rules! read_value {
    ($iter:expr, ( $($t:tt),* )) => {
        ( $(read_value!($iter, $t)),* )
    };
    ($iter:expr, [ $t:tt ; $len:expr ]) => {
        (0..$len).map(|_| read_value!($iter, $t)).collect::<Vec<_>>()
    };
    ($iter:expr, chars) => {
        read_value!($iter, String).chars().collect::<Vec<char>>()
    };
    ($iter:expr, bytes) => {
        read_value!($iter, String).bytes().collect::<Vec<u8>>()
    };
    ($iter:expr, usize1) => {
        read_value!($iter, usize) - 1
    };
    ($iter:expr, $t:ty) => {
        $iter.next().unwrap().parse::<$t>().expect("Parse error")
    };
}
// ---------- end input macro ----------
// ---------- begin Lazy Segment Tree ----------
pub trait TE {
    type T: Clone;
    type E: Clone;
    fn fold(&self, l: &Self::T, r: &Self::T) -> Self::T;
    fn eval(&self, x: &Self::T, f: &Self::E) -> Self::T;
    fn merge(&self, g: &Self::E, h: &Self::E) -> Self::E;
    fn e(&self) -> Self::T;
    fn id(&self) -> Self::E;
}

pub struct LazySegmentTree<R: TE> {
    n: usize,
    size: usize,
    bit: u32,
    op: R,
    data: Vec<(R::T, R::E)>,
}

impl<R: TE> LazySegmentTree<R> {
    pub fn new(n: usize, op: R) -> Self {
        assert!(n > 0);
        let size = n.next_power_of_two();
        let bit = size.trailing_zeros();
        let data = vec![(op.e(), op.id()); 2 * size];
        Self {
            n,
            size,
            bit,
            op,
            data,
        }
    }
    pub fn build<I>(init: I, n: usize, op: R) -> Self
    where
        I: Iterator<Item = R::T>,
    {
        let mut seg = Self::new(n, op);
        for (data, ini) in seg.data[seg.size..].iter_mut().zip(init) {
            data.0 = ini;
        }
        for i in (1..seg.size).rev() {
            seg.pull(i);
        }
        seg
    }
    pub fn update(&mut self, l: usize, r: usize, f: R::E) {
        assert!(l <= r && r <= self.n);
        if l == r {
            return;
        }
        self.push_range(l, r);
        let mut s = l + self.size;
        let mut t = r + self.size;
        while s < t {
            if s & 1 == 1 {
                self.apply(s, &f);
                s += 1;
            }
            if t & 1 == 1 {
                t -= 1;
                self.apply(t, &f);
            }
            s >>= 1;
            t >>= 1;
        }
        let l = l + self.size;
        let r = r + self.size;
        for k in 1..=self.bit {
            if (l >> k) << k != l {
                self.pull(l >> k);
            }
            if (r >> k) << k != r {
                self.pull((r - 1) >> k);
            }
        }
    }
    pub fn find(&mut self, l: usize, r: usize) -> R::T {
        assert!(l <= r && r <= self.n);
        if l == r {
            return self.op.e();
        }
        self.push_range(l, r);
        let mut l = l + self.size;
        let mut r = r + self.size;
        let mut p = self.op.e();
        let mut q = self.op.e();
        while l < r {
            if l & 1 == 1 {
                p = self.op.fold(&p, &self.data[l].0);
                l += 1;
            }
            if r & 1 == 1 {
                r -= 1;
                q = self.op.fold(&self.data[r].0, &q);
            }
            l >>= 1;
            r >>= 1;
        }
        self.op.fold(&p, &q)
    }
    pub fn set_at(&mut self, x: usize, v: R::T) {
        assert!(x < self.n);
        let x = x + self.size;
        for k in (1..=self.bit).rev() {
            self.push(x >> k);
        }
        self.data[x].0 = v;
        for k in 1..=self.bit {
            self.pull(x >> k);
        }
    }
    fn push_range(&mut self, l: usize, r: usize) {
        let l = l + self.size;
        let r = r + self.size;
        for k in (1..(self.bit + 1)).rev() {
            if (l >> k) << k != l {
                self.push(l >> k);
            }
            if (r >> k) << k != r {
                self.push((r - 1) >> k);
            }
        }
    }
    fn apply(&mut self, x: usize, f: &R::E) {
        self.data[x].0 = self.op.eval(&self.data[x].0, f);
        self.data[x].1 = self.op.merge(&self.data[x].1, f);
    }
    fn push(&mut self, x: usize) {
        let f = std::mem::replace(&mut self.data[x].1, self.op.id());
        self.apply(2 * x, &f);
        self.apply(2 * x + 1, &f);
    }
    fn pull(&mut self, x: usize) {
        self.data[x].0 = self.op.fold(&self.data[2 * x].0, &self.data[2 * x + 1].0);
    }
}
// ---------- end Lazy Segment Tree ----------

use std::ops::*;

// ---------- begin trait ----------
pub trait Zero: Sized + Add<Self, Output = Self> {
    fn zero() -> Self;
    fn is_zero(&self) -> bool;
}

pub trait One: Sized + Mul<Self, Output = Self> {
    fn one() -> Self;
    fn is_one(&self) -> bool;
}

pub trait SemiRing: Zero + One {}

pub trait Ring: SemiRing + Sub<Output = Self> + Neg<Output = Self> {}

pub trait Field: Ring + Div<Output = Self> {}

impl<T> SemiRing for T where T: Zero + One {}

impl<T> Ring for T where T: SemiRing + Sub<Output = Self> + Neg<Output = Self> {}

impl<T> Field for T where T: Ring + Div<Output = Self> {}
// ---------- end trait ----------
// ---------- begin modint ----------
pub const fn pow_mod(mut r: u32, mut n: u32, m: u32) -> u32 {
    let mut t = 1;
    while n > 0 {
        if n & 1 == 1 {
            t = (t as u64 * r as u64 % m as u64) as u32;
        }
        r = (r as u64 * r as u64 % m as u64) as u32;
        n >>= 1;
    }
    t
}

pub const fn primitive_root(p: u32) -> u32 {
    let mut m = p - 1;
    let mut f = [1; 30];
    let mut k = 0;
    let mut d = 2;
    while d * d <= m {
        if m % d == 0 {
            f[k] = d;
            k += 1;
        }
        while m % d == 0 {
            m /= d;
        }
        d += 1;
    }
    if m > 1 {
        f[k] = m;
        k += 1;
    }
    let mut g = 1;
    while g < p {
        let mut ok = true;
        let mut i = 0;
        while i < k {
            ok &= pow_mod(g, (p - 1) / f[i], p) > 1;
            i += 1;
        }
        if ok {
            break;
        }
        g += 1;
    }
    g
}

pub const fn is_prime(n: u32) -> bool {
    if n <= 1 {
        return false;
    }
    let mut d = 2;
    while d * d <= n {
        if n % d == 0 {
            return false;
        }
        d += 1;
    }
    true
}

#[derive(Clone, Copy, PartialEq, Eq)]
pub struct ModInt<const M: u32>(u32);

impl<const M: u32> ModInt<{ M }> {
    const REM: u32 = {
        let mut t = 1u32;
        let mut s = !M + 1;
        let mut n = !0u32 >> 2;
        while n > 0 {
            if n & 1 == 1 {
                t = t.wrapping_mul(s);
            }
            s = s.wrapping_mul(s);
            n >>= 1;
        }
        t
    };
    const INI: u64 = ((1u128 << 64) % M as u128) as u64;
    const IS_PRIME: () = assert!(is_prime(M));
    const PRIMITIVE_ROOT: u32 = primitive_root(M);
    const ORDER: usize = 1 << (M - 1).trailing_zeros();
    const fn reduce(x: u64) -> u32 {
        let _ = Self::IS_PRIME;
        let b = (x as u32 * Self::REM) as u64;
        let t = x + b * M as u64;
        let mut c = (t >> 32) as u32;
        if c >= M {
            c -= M;
        }
        c as u32
    }
    const fn multiply(a: u32, b: u32) -> u32 {
        Self::reduce(a as u64 * b as u64)
    }
    pub const fn new(v: u32) -> Self {
        assert!(v < M);
        Self(Self::reduce(v as u64 * Self::INI))
    }
    pub const fn const_mul(&self, rhs: Self) -> Self {
        Self(Self::multiply(self.0, rhs.0))
    }
    pub const fn pow(&self, mut n: u64) -> Self {
        let mut t = Self::new(1);
        let mut r = *self;
        while n > 0 {
            if n & 1 == 1 {
                t = t.const_mul(r);
            }
            r = r.const_mul(r);
            n >>= 1;
        }
        t
    }
    pub const fn inv(&self) -> Self {
        assert!(self.0 != 0);
        self.pow(M as u64 - 2)
    }
    pub const fn get(&self) -> u32 {
        Self::reduce(self.0 as u64)
    }
    pub const fn zero() -> Self {
        Self::new(0)
    }
    pub const fn one() -> Self {
        Self::new(1)
    }
}

impl<const M: u32> Add for ModInt<{ M }> {
    type Output = Self;
    fn add(self, rhs: Self) -> Self::Output {
        let mut v = self.0 + rhs.0;
        if v >= M {
            v -= M;
        }
        Self(v)
    }
}

impl<const M: u32> Sub for ModInt<{ M }> {
    type Output = Self;
    fn sub(self, rhs: Self) -> Self::Output {
        let mut v = self.0 - rhs.0;
        if self.0 < rhs.0 {
            v += M;
        }
        Self(v)
    }
}

impl<const M: u32> Mul for ModInt<{ M }> {
    type Output = Self;
    fn mul(self, rhs: Self) -> Self::Output {
        self.const_mul(rhs)
    }
}

impl<const M: u32> Div for ModInt<{ M }> {
    type Output = Self;
    fn div(self, rhs: Self) -> Self::Output {
        self * rhs.inv()
    }
}

impl<const M: u32> AddAssign for ModInt<{ M }> {
    fn add_assign(&mut self, rhs: Self) {
        *self = *self + rhs;
    }
}

impl<const M: u32> SubAssign for ModInt<{ M }> {
    fn sub_assign(&mut self, rhs: Self) {
        *self = *self - rhs;
    }
}

impl<const M: u32> MulAssign for ModInt<{ M }> {
    fn mul_assign(&mut self, rhs: Self) {
        *self = *self * rhs;
    }
}

impl<const M: u32> DivAssign for ModInt<{ M }> {
    fn div_assign(&mut self, rhs: Self) {
        *self = *self / rhs;
    }
}

impl<const M: u32> Neg for ModInt<{ M }> {
    type Output = Self;
    fn neg(self) -> Self::Output {
        if self.0 == 0 {
            self
        } else {
            Self(M - self.0)
        }
    }
}

impl<const M: u32> std::fmt::Display for ModInt<{ M }> {
    fn fmt<'a>(&self, f: &mut std::fmt::Formatter<'a>) -> std::fmt::Result {
        write!(f, "{}", self.get())
    }
}

impl<const M: u32> std::fmt::Debug for ModInt<{ M }> {
    fn fmt<'a>(&self, f: &mut std::fmt::Formatter<'a>) -> std::fmt::Result {
        write!(f, "{}", self.get())
    }
}

impl<const M: u32> std::str::FromStr for ModInt<{ M }> {
    type Err = std::num::ParseIntError;
    fn from_str(s: &str) -> Result<Self, Self::Err> {
        let val = s.parse::<u32>()?;
        Ok(ModInt::new(val))
    }
}

impl<const M: u32> From<usize> for ModInt<{ M }> {
    fn from(val: usize) -> ModInt<{ M }> {
        ModInt::new((val % M as usize) as u32)
    }
}
// ---------- end modint ----------
// ---------- begin precalc ----------
pub struct Precalc<const MOD: u32> {
    fact: Vec<ModInt<MOD>>,
    ifact: Vec<ModInt<MOD>>,
    inv: Vec<ModInt<MOD>>,
}

impl<const MOD: u32> Precalc<MOD> {
    pub fn new(size: usize) -> Self {
        let mut fact = vec![ModInt::one(); size + 1];
        let mut ifact = vec![ModInt::one(); size + 1];
        let mut inv = vec![ModInt::one(); size + 1];
        for i in 2..=size {
            fact[i] = fact[i - 1] * ModInt::from(i);
        }
        ifact[size] = fact[size].inv();
        for i in (2..=size).rev() {
            inv[i] = ifact[i] * fact[i - 1];
            ifact[i - 1] = ifact[i] * ModInt::from(i);
        }
        Self { fact, ifact, inv }
    }
    pub fn fact(&self, n: usize) -> ModInt<MOD> {
        self.fact[n]
    }
    pub fn ifact(&self, n: usize) -> ModInt<MOD> {
        self.ifact[n]
    }
    pub fn inv(&self, n: usize) -> ModInt<MOD> {
        assert!(0 < n);
        self.inv[n]
    }
    pub fn perm(&self, n: usize, k: usize) -> ModInt<MOD> {
        if k > n {
            return ModInt::zero();
        }
        self.fact[n] * self.ifact[n - k]
    }
    pub fn binom(&self, n: usize, k: usize) -> ModInt<MOD> {
        if n < k {
            return ModInt::zero();
        }
        self.fact[n] * self.ifact[k] * self.ifact[n - k]
    }
}
// ---------- end precalc ----------

impl<const M: u32> Zero for ModInt<{ M }> {
    fn zero() -> Self {
        Self::zero()
    }
    fn is_zero(&self) -> bool {
        self.0 == 0
    }
}

impl<const M: u32> One for ModInt<{ M }> {
    fn one() -> Self {
        Self::one()
    }
    fn is_one(&self) -> bool {
        self.get() == 1
    }
}

// ---------- begin array op ----------

struct NTTPrecalc<const M: u32> {
    sum_e: [ModInt<{ M }>; 30],
    sum_ie: [ModInt<{ M }>; 30],
}

impl<const M: u32> NTTPrecalc<{ M }> {
    const fn new() -> Self {
        let cnt2 = (M - 1).trailing_zeros() as usize;
        let root = ModInt::new(ModInt::<{ M }>::PRIMITIVE_ROOT);
        let zeta = root.pow((M - 1) as u64 >> cnt2);
        let mut es = [ModInt::zero(); 30];
        let mut ies = [ModInt::zero(); 30];
        let mut sum_e = [ModInt::zero(); 30];
        let mut sum_ie = [ModInt::zero(); 30];
        let mut e = zeta;
        let mut ie = e.inv();
        let mut i = cnt2;
        while i >= 2 {
            es[i - 2] = e;
            ies[i - 2] = ie;
            e = e.const_mul(e);
            ie = ie.const_mul(ie);
            i -= 1;
        }
        let mut now = ModInt::one();
        let mut inow = ModInt::one();
        let mut i = 0;
        while i < cnt2 - 1 {
            sum_e[i] = es[i].const_mul(now);
            sum_ie[i] = ies[i].const_mul(inow);
            now = ies[i].const_mul(now);
            inow = es[i].const_mul(inow);
            i += 1;
        }
        Self { sum_e, sum_ie }
    }
}

struct NTTPrecalcHelper<const MOD: u32>;
impl<const MOD: u32> NTTPrecalcHelper<MOD> {
    const A: NTTPrecalc<MOD> = NTTPrecalc::new();
}

pub trait ArrayAdd {
    type Item;
    fn add(&self, rhs: &[Self::Item]) -> Vec<Self::Item>;
}

impl<T> ArrayAdd for [T]
where
    T: Zero + Copy,
{
    type Item = T;
    fn add(&self, rhs: &[Self::Item]) -> Vec<Self::Item> {
        let mut c = vec![T::zero(); self.len().max(rhs.len())];
        c[..self.len()].copy_from_slice(self);
        c.add_assign(rhs);
        c
    }
}

pub trait ArrayAddAssign {
    type Item;
    fn add_assign(&mut self, rhs: &[Self::Item]);
}

impl<T> ArrayAddAssign for [T]
where
    T: Add<Output = T> + Copy,
{
    type Item = T;
    fn add_assign(&mut self, rhs: &[Self::Item]) {
        assert!(self.len() >= rhs.len());
        self.iter_mut().zip(rhs).for_each(|(x, a)| *x = *x + *a);
    }
}

impl<T> ArrayAddAssign for Vec<T>
where
    T: Zero + Add<Output = T> + Copy,
{
    type Item = T;
    fn add_assign(&mut self, rhs: &[Self::Item]) {
        if self.len() < rhs.len() {
            self.resize(rhs.len(), T::zero());
        }
        self.as_mut_slice().add_assign(rhs);
    }
}

pub trait ArraySub {
    type Item;
    fn sub(&self, rhs: &[Self::Item]) -> Vec<Self::Item>;
}

impl<T> ArraySub for [T]
where
    T: Zero + Sub<Output = T> + Copy,
{
    type Item = T;
    fn sub(&self, rhs: &[Self::Item]) -> Vec<Self::Item> {
        let mut c = vec![T::zero(); self.len().max(rhs.len())];
        c[..self.len()].copy_from_slice(self);
        c.sub_assign(rhs);
        c
    }
}

pub trait ArraySubAssign {
    type Item;
    fn sub_assign(&mut self, rhs: &[Self::Item]);
}

impl<T> ArraySubAssign for [T]
where
    T: Sub<Output = T> + Copy,
{
    type Item = T;
    fn sub_assign(&mut self, rhs: &[Self::Item]) {
        assert!(self.len() >= rhs.len());
        self.iter_mut().zip(rhs).for_each(|(x, a)| *x = *x - *a);
    }
}

impl<T> ArraySubAssign for Vec<T>
where
    T: Zero + Sub<Output = T> + Copy,
{
    type Item = T;
    fn sub_assign(&mut self, rhs: &[Self::Item]) {
        if self.len() < rhs.len() {
            self.resize(rhs.len(), T::zero());
        }
        self.as_mut_slice().sub_assign(rhs);
    }
}

pub trait ArrayDot {
    type Item;
    fn dot(&self, rhs: &[Self::Item]) -> Vec<Self::Item>;
}

impl<T> ArrayDot for [T]
where
    T: Mul<Output = T> + Copy,
{
    type Item = T;
    fn dot(&self, rhs: &[Self::Item]) -> Vec<Self::Item> {
        assert!(self.len() == rhs.len());
        self.iter().zip(rhs).map(|p| *p.0 * *p.1).collect()
    }
}

pub trait ArrayDotAssign {
    type Item;
    fn dot_assign(&mut self, rhs: &[Self::Item]);
}

impl<T> ArrayDotAssign for [T]
where
    T: MulAssign + Copy,
{
    type Item = T;
    fn dot_assign(&mut self, rhs: &[Self::Item]) {
        assert!(self.len() == rhs.len());
        self.iter_mut().zip(rhs).for_each(|(x, a)| *x *= *a);
    }
}

pub trait ArrayMul {
    type Item;
    fn mul(&self, rhs: &[Self::Item]) -> Vec<Self::Item>;
}

impl<T> ArrayMul for [T]
where
    T: Zero + One + Copy,
{
    type Item = T;
    fn mul(&self, rhs: &[Self::Item]) -> Vec<Self::Item> {
        if self.is_empty() || rhs.is_empty() {
            return vec![];
        }
        let mut res = vec![T::zero(); self.len() + rhs.len() - 1];
        for (i, a) in self.iter().enumerate() {
            for (res, b) in res[i..].iter_mut().zip(rhs.iter()) {
                *res = *res + *a * *b;
            }
        }
        res
    }
}

// transform でlen=1を指定すればNTTになる
pub trait ArrayConvolution {
    type Item;
    fn transform(&mut self, len: usize);
    fn inverse_transform(&mut self, len: usize);
    fn convolution(&self, rhs: &[Self::Item]) -> Vec<Self::Item>;
}

impl<const M: u32> ArrayConvolution for [ModInt<{ M }>] {
    type Item = ModInt<{ M }>;
    fn transform(&mut self, len: usize) {
        let f = self;
        let n = f.len();
        let k = (n / len).trailing_zeros() as usize;
        assert!(len << k == n);
        assert!(k <= ModInt::<{ M }>::ORDER);
        let pre = &NTTPrecalcHelper::<{ M }>::A;
        for ph in 1..=k {
            let p = len << (k - ph);
            let mut now = ModInt::one();
            for (i, f) in f.chunks_exact_mut(2 * p).enumerate() {
                let (x, y) = f.split_at_mut(p);
                for (x, y) in x.iter_mut().zip(y.iter_mut()) {
                    let l = *x;
                    let r = *y * now;
                    *x = l + r;
                    *y = l - r;
                }
                now *= pre.sum_e[(!i).trailing_zeros() as usize];
            }
        }
    }
    fn inverse_transform(&mut self, len: usize) {
        let f = self;
        let n = f.len();
        let k = (n / len).trailing_zeros() as usize;
        assert!(len << k == n);
        assert!(k <= ModInt::<{ M }>::ORDER);
        let pre = &NTTPrecalcHelper::<{ M }>::A;
        for ph in (1..=k).rev() {
            let p = len << (k - ph);
            let mut inow = ModInt::one();
            for (i, f) in f.chunks_exact_mut(2 * p).enumerate() {
                let (x, y) = f.split_at_mut(p);
                for (x, y) in x.iter_mut().zip(y.iter_mut()) {
                    let l = *x;
                    let r = *y;
                    *x = l + r;
                    *y = (l - r) * inow;
                }
                inow *= pre.sum_ie[(!i).trailing_zeros() as usize];
            }
        }
        let ik = ModInt::new(2).inv().pow(k as u64);
        for f in f.iter_mut() {
            *f *= ik;
        }
    }
    fn convolution(&self, rhs: &[Self::Item]) -> Vec<Self::Item> {
        if self.len().min(rhs.len()) <= 32 {
            return self.mul(rhs);
        }
        const PARAM: usize = 10;
        let size = self.len() + rhs.len() - 1;
        let mut k = 0;
        while (size + (1 << k) - 1) >> k > PARAM {
            k += 1;
        }
        let len = (size + (1 << k) - 1) >> k;
        let mut f = vec![ModInt::zero(); len << k];
        let mut g = vec![ModInt::zero(); len << k];
        f[..self.len()].copy_from_slice(self);
        g[..rhs.len()].copy_from_slice(rhs);
        f.transform(len);
        g.transform(len);
        let mut buf = [ModInt::zero(); 2 * PARAM - 1];
        let buf = &mut buf[..(2 * len - 1)];
        let pre = &NTTPrecalcHelper::<{ M }>::A;
        let mut now = ModInt::one();
        for (i, (f, g)) in f
            .chunks_exact_mut(2 * len)
            .zip(g.chunks_exact(2 * len))
            .enumerate()
        {
            let mut r = now;
            for (f, g) in f.chunks_exact_mut(len).zip(g.chunks_exact(len)) {
                buf.fill(ModInt::zero());
                for (i, f) in f.iter().enumerate() {
                    for (buf, g) in buf[i..].iter_mut().zip(g.iter()) {
                        *buf = *buf + *f * *g;
                    }
                }
                f.copy_from_slice(&buf[..len]);
                for (f, buf) in f.iter_mut().zip(buf[len..].iter()) {
                    *f = *f + r * *buf;
                }
                r = -r;
            }
            now *= pre.sum_e[(!i).trailing_zeros() as usize];
        }
        f.inverse_transform(len);
        f.truncate(self.len() + rhs.len() - 1);
        f
    }
}
// ---------- end array op ----------