NTT Convolution
(convolution/ntt.hpp)

• View this file on GitHub
• Last update: 2023-10-31 00:17:11+09:00
• Include: #include "convolution/ntt.hpp"

説明

NTT friendlyな素数における $O(N\log N)$ 時間での畳み込み。

ntt_info

NTTをするために必要なデータを格納している。

butterfly(std::vector &a)

配列 $a$ をNTTする。 $a$ の配列の大きさは $2$ 冪でのみ動作する。

butterfly_inv(std::vector &a)

配列 $a$ をinverse NTTする。 $a$ の配列の大きさは $2$ 冪でのみ動作する。 inverse NTT後に割る $N$ を行う。

convolution(std::vector a, std::vector b)

$a$ と $b$ を畳み込み、その配列を返す。 $O(N\log N)$

Depends on

• math/internal_math.hpp
• modint/base.hpp

Required by

• Arbitrary Convolution (convolution/arbitrary_ntt.hpp)
• Convolution $\pmod{2^{64}}$ (convolution/convolution_mod_2_64.hpp)

Verified with

• test/convolution/Convolution.test.cpp
• test/convolution/Convolution_Mod_1000000007.test.cpp
• test/convolution/Convolution_Mod_2_64.test.cpp
• test/math/Berunoulli_Number.test.cpp
• test/math/Catalan_Convolution.test.cpp
• test/math/Kth_term_of_Linearly_Recurrent_Sequence.test.cpp
• test/math/Partition_Function_FPS.test.cpp
• test/math/Partition_Function_Pentagonal.test.cpp
• test/math/Stirling_Number_of_the_First_Kind.test.cpp
• test/math/Stirling_Number_of_the_Second_Kind.test.cpp
• test/math/sharp_p_subset_sum.test.cpp
• test/polynomial/Composition_of_Formal_Power_Series.test.cpp
• test/polynomial/Compositional_Inverse_of_Formal_Power_Series.test.cpp
• test/polynomial/Division_of_Polynomials.test.cpp
• test/polynomial/Exp_of_Formal_Power_Series.test.cpp
• test/polynomial/Inv_of_Formal_Power_Series.test.cpp
• test/polynomial/Log_of_Formal_Power_Series.test.cpp
• test/polynomial/Multipoint_Evaluation.test.cpp
• test/polynomial/Polynomial_Interpolation.test.cpp
• test/polynomial/Polynomial_Taylor_Shift.test.cpp
• test/polynomial/Pow_of_Formal_Power_Series.test.cpp
• test/polynomial/Product_of_Polynomial_Sequence.test.cpp
• test/polynomial/Sqrt_of_Formal_Power_Series.test.cpp
• test/polynomial/Sqrt_of_Formal_Power_Series_Sparse.test.cpp
• test/tree/Frequency_Table_of_Tree_Distance_MODE_0.test.cpp
• test/tree/Frequency_Table_of_Tree_Distance_MODE_1.test.cpp
• test/tree/Frequency_Table_of_Tree_Distance_MODE_2.test.cpp
• test/yuki/yuki_1145.test.cpp
• test/yuki/yuki_1302.test.cpp
• test/yuki/yuki_1796.test.cpp

Code

#pragma once

#include <algorithm>
#include <array>
#include <bit>
#include <cassert>
#include <vector>

#include "../math/internal_math.hpp"
#include "../modint/base.hpp"

namespace ebi {

namespace internal {

template <Modint mint, int g = internal::primitive_root<mint::mod()>>
struct ntt_info {
    static constexpr int rank2 =
        std::countr_zero((unsigned int)(mint::mod() - 1));

    std::array<mint, rank2 + 1> root, inv_root;

    ntt_info() {
        root[rank2] = mint(g).pow((mint::mod() - 1) >> rank2);
        inv_root[rank2] = root[rank2].inv();
        for (int i = rank2 - 1; i >= 0; i--) {
            root[i] = root[i + 1] * root[i + 1];
            inv_root[i] = inv_root[i + 1] * inv_root[i + 1];
        }
    }
};

template <Modint mint> void butterfly(std::vector<mint>& a) {
    static const ntt_info<mint> info;
    int n = int(a.size());
    int bit_size = std::countr_zero(a.size());
    assert(n == (int)std::bit_ceil(a.size()));

    // bit reverse
    for (int i = 0, j = 1; j < n - 1; j++) {
        for (int k = n >> 1; k > (i ^= k); k >>= 1)
            ;
        if (j < i) {
            std::swap(a[i], a[j]);
        }
    }

    for (int bit = 0; bit < bit_size; bit++) {
        for (int i = 0; i < n / (1 << (bit + 1)); i++) {
            mint zeta1 = 1;
            mint zeta2 = info.root[1];
            for (int j = 0; j < (1 << bit); j++) {
                int idx = i * (1 << (bit + 1)) + j;
                int jdx = idx + (1 << bit);
                mint p1 = a[idx];
                mint p2 = a[jdx];
                a[idx] = p1 + zeta1 * p2;
                a[jdx] = p1 + zeta2 * p2;
                zeta1 *= info.root[bit + 1];
                zeta2 *= info.root[bit + 1];
            }
        }
    }
}

template <Modint mint> void butterfly_inv(std::vector<mint>& a) {
    static const ntt_info<mint> info;
    int n = int(a.size());
    int bit_size = std::countr_zero(a.size());
    assert(n == (int)std::bit_ceil(a.size()));

    // bit reverse
    for (int i = 0, j = 1; j < n - 1; j++) {
        for (int k = n >> 1; k > (i ^= k); k >>= 1)
            ;
        if (j < i) {
            std::swap(a[i], a[j]);
        }
    }

    for (int bit = 0; bit < bit_size; bit++) {
        for (int i = 0; i < n / (1 << (bit + 1)); i++) {
            mint zeta1 = 1;
            mint zeta2 = info.inv_root[1];
            for (int j = 0; j < (1 << bit); j++) {
                int idx = i * (1 << (bit + 1)) + j;
                int jdx = idx + (1 << bit);
                mint p1 = a[idx];
                mint p2 = a[jdx];
                a[idx] = p1 + zeta1 * p2;
                a[jdx] = p1 + zeta2 * p2;
                zeta1 *= info.inv_root[bit + 1];
                zeta2 *= info.inv_root[bit + 1];
            }
        }
    }
    mint inv_n = mint(n).inv();
    for (int i = 0; i < n; i++) {
        a[i] *= inv_n;
    }
}

}  // namespace internal

template <Modint mint>
std::vector<mint> convolution_naive(const std::vector<mint>& f,
                                    const std::vector<mint>& g) {
    if (f.empty() || g.empty()) return {};
    int n = int(f.size()), m = int(g.size());
    std::vector<mint> c(n + m - 1);
    if (n < m) {
        for (int j = 0; j < m; j++) {
            for (int i = 0; i < n; i++) {
                c[i + j] += f[i] * g[j];
            }
        }
    } else {
        for (int i = 0; i < n; i++) {
            for (int j = 0; j < m; j++) {
                c[i + j] += f[i] * g[j];
            }
        }
    }
    return c;
}

template <Modint mint>
std::vector<mint> convolution(const std::vector<mint>& f,
                              const std::vector<mint>& g) {
    if (f.empty() || g.empty()) return {};
    if (std::min(f.size(), g.size()) < 60) return convolution_naive(f, g);
    int n = std::bit_ceil(f.size() + g.size() - 1);
    std::vector<mint> a(n), b(n);
    std::copy(f.begin(), f.end(), a.begin());
    std::copy(g.begin(), g.end(), b.begin());
    internal::butterfly(a);
    internal::butterfly(b);
    for (int i = 0; i < n; i++) {
        a[i] *= b[i];
    }
    internal::butterfly_inv(a);
    a.resize(f.size() + g.size() - 1);
    return a;
}

}  // namespace ebi

#line 2 "convolution/ntt.hpp"

#include <algorithm>
#include <array>
#include <bit>
#include <cassert>
#include <vector>

#line 2 "math/internal_math.hpp"

#line 4 "math/internal_math.hpp"

namespace ebi {

namespace internal {

constexpr int primitive_root_constexpr(int m) {
    if (m == 2) return 1;
    if (m == 167772161) return 3;
    if (m == 469762049) return 3;
    if (m == 754974721) return 11;
    if (m == 998244353) return 3;
    if (m == 880803841) return 26;
    if (m == 924844033) return 5;
    return -1;
}
template <int m> constexpr int primitive_root = primitive_root_constexpr(m);

}  // namespace internal

}  // namespace ebi
#line 2 "modint/base.hpp"

#include <concepts>
#include <iostream>
#include <utility>

namespace ebi {

template <class T>
concept Modint = requires(T a, T b) {
    a + b;
    a - b;
    a * b;
    a / b;
    a.inv();
    a.val();
    a.pow(std::declval<long long>());
    T::mod();
};

template <Modint mint> std::istream &operator>>(std::istream &os, mint &a) {
    long long x;
    os >> x;
    a = x;
    return os;
}

template <Modint mint>
std::ostream &operator<<(std::ostream &os, const mint &a) {
    return os << a.val();
}

}  // namespace ebi
#line 11 "convolution/ntt.hpp"

namespace ebi {

namespace internal {

template <Modint mint, int g = internal::primitive_root<mint::mod()>>
struct ntt_info {
    static constexpr int rank2 =
        std::countr_zero((unsigned int)(mint::mod() - 1));

    std::array<mint, rank2 + 1> root, inv_root;

    ntt_info() {
        root[rank2] = mint(g).pow((mint::mod() - 1) >> rank2);
        inv_root[rank2] = root[rank2].inv();
        for (int i = rank2 - 1; i >= 0; i--) {
            root[i] = root[i + 1] * root[i + 1];
            inv_root[i] = inv_root[i + 1] * inv_root[i + 1];
        }
    }
};

template <Modint mint> void butterfly(std::vector<mint>& a) {
    static const ntt_info<mint> info;
    int n = int(a.size());
    int bit_size = std::countr_zero(a.size());
    assert(n == (int)std::bit_ceil(a.size()));

    // bit reverse
    for (int i = 0, j = 1; j < n - 1; j++) {
        for (int k = n >> 1; k > (i ^= k); k >>= 1)
            ;
        if (j < i) {
            std::swap(a[i], a[j]);
        }
    }

    for (int bit = 0; bit < bit_size; bit++) {
        for (int i = 0; i < n / (1 << (bit + 1)); i++) {
            mint zeta1 = 1;
            mint zeta2 = info.root[1];
            for (int j = 0; j < (1 << bit); j++) {
                int idx = i * (1 << (bit + 1)) + j;
                int jdx = idx + (1 << bit);
                mint p1 = a[idx];
                mint p2 = a[jdx];
                a[idx] = p1 + zeta1 * p2;
                a[jdx] = p1 + zeta2 * p2;
                zeta1 *= info.root[bit + 1];
                zeta2 *= info.root[bit + 1];
            }
        }
    }
}

template <Modint mint> void butterfly_inv(std::vector<mint>& a) {
    static const ntt_info<mint> info;
    int n = int(a.size());
    int bit_size = std::countr_zero(a.size());
    assert(n == (int)std::bit_ceil(a.size()));

    // bit reverse
    for (int i = 0, j = 1; j < n - 1; j++) {
        for (int k = n >> 1; k > (i ^= k); k >>= 1)
            ;
        if (j < i) {
            std::swap(a[i], a[j]);
        }
    }

    for (int bit = 0; bit < bit_size; bit++) {
        for (int i = 0; i < n / (1 << (bit + 1)); i++) {
            mint zeta1 = 1;
            mint zeta2 = info.inv_root[1];
            for (int j = 0; j < (1 << bit); j++) {
                int idx = i * (1 << (bit + 1)) + j;
                int jdx = idx + (1 << bit);
                mint p1 = a[idx];
                mint p2 = a[jdx];
                a[idx] = p1 + zeta1 * p2;
                a[jdx] = p1 + zeta2 * p2;
                zeta1 *= info.inv_root[bit + 1];
                zeta2 *= info.inv_root[bit + 1];
            }
        }
    }
    mint inv_n = mint(n).inv();
    for (int i = 0; i < n; i++) {
        a[i] *= inv_n;
    }
}

}  // namespace internal

template <Modint mint>
std::vector<mint> convolution_naive(const std::vector<mint>& f,
                                    const std::vector<mint>& g) {
    if (f.empty() || g.empty()) return {};
    int n = int(f.size()), m = int(g.size());
    std::vector<mint> c(n + m - 1);
    if (n < m) {
        for (int j = 0; j < m; j++) {
            for (int i = 0; i < n; i++) {
                c[i + j] += f[i] * g[j];
            }
        }
    } else {
        for (int i = 0; i < n; i++) {
            for (int j = 0; j < m; j++) {
                c[i + j] += f[i] * g[j];
            }
        }
    }
    return c;
}

template <Modint mint>
std::vector<mint> convolution(const std::vector<mint>& f,
                              const std::vector<mint>& g) {
    if (f.empty() || g.empty()) return {};
    if (std::min(f.size(), g.size()) < 60) return convolution_naive(f, g);
    int n = std::bit_ceil(f.size() + g.size() - 1);
    std::vector<mint> a(n), b(n);
    std::copy(f.begin(), f.end(), a.begin());
    std::copy(g.begin(), g.end(), b.begin());
    internal::butterfly(a);
    internal::butterfly(b);
    for (int i = 0; i < n; i++) {
        a[i] *= b[i];
    }
    internal::butterfly_inv(a);
    a.resize(f.size() + g.size() - 1);
    return a;
}

}  // namespace ebi

Back to top page