/// @file grotto/fixedpoint.hpp
/// @brief
/// @details
/// @author Ryan Henry <ryan.henry@ucalgary.ca>
/// @copyright Copyright (c) 2019-2023 Ryan Henry and others
/// @license Released under a GNU General Public v2.0 (GPLv2) license;
///          see [LICENSE.md](@ref GPLv2) for details.

#ifndef LIBDPF_INCLUDE_DPF_FIXEDPOINT_HPP__
#define LIBDPF_INCLUDE_DPF_FIXEDPOINT_HPP__

#include "hedley/hedley.h"
#include <portable-snippets/exact-int/exact-int.h>
#include <portable-snippets/builtin/builtin.h>

#include <cstddef>
#include <cstdint>
#include <cinttypes>
#include <limits>
#include <type_traits>
#include <functional>
#include <array>
#include <stdexcept>
#include <utility>

#include <cmath>
#include <iostream>

#include "dpf/utils.hpp"
#include "dpf/leaf_arithmetic.hpp"
#include "dpf/uint256_t.hpp"

#define GROTTO_FIXED_DEFAULT_INTEGRAL_REPRESENTATION psnip_uint64_t

namespace grotto
{

namespace detail
{

/// @brief Integer value of an already-rounded finite double, as a 256-bit word.
///        Values that do not fit saturate to all-ones.
inline uint256_t uint256_from_rounded_double(double rounded) noexcept
{
    if (!(rounded > 0.0) || !std::isfinite(rounded))
    {
        return uint256_t{0};
    }
    int exp = 0;
    const double frac = std::frexp(rounded, &exp);
    if (exp <= 0)
    {
        return uint256_t{0};
    }
    if (exp > 256)
    {
        return ~uint256_t{0};
    }
    constexpr int mant_bits = 53;
    const auto mant = static_cast<std::uint64_t>(std::ldexp(frac, mant_bits));
    const int place = exp - mant_bits;
    simde_uint128 chunk = mant;
    int bit = place;
    if (bit < 0)
    {
        chunk >>= static_cast<unsigned>(-bit);
        bit = 0;
    }
    uint128_t lower{0};
    uint128_t upper{0};
    if (bit < 128)
    {
        const simde_uint128 lowbits = chunk << static_cast<unsigned>(bit);
        lower = uint128_t{static_cast<std::uint64_t>(lowbits >> 64),
            static_cast<std::uint64_t>(lowbits)};
        if (bit > 128 - mant_bits)
        {
            const simde_uint128 hibits = chunk >> static_cast<unsigned>(128 - bit);
            upper = uint128_t{static_cast<std::uint64_t>(hibits >> 64),
                static_cast<std::uint64_t>(hibits)};
        }
    }
    else
    {
        const simde_uint128 hibits = chunk << static_cast<unsigned>(bit - 128);
        upper = uint128_t{static_cast<std::uint64_t>(hibits >> 64),
            static_cast<std::uint64_t>(hibits)};
    }
    return uint256_t{upper, lower};
}

template <typename To, typename From, typename = void>
struct is_static_castable : std::false_type {};

template <typename To, typename From>
struct is_static_castable<To, From,
    std::void_t<decltype(static_cast<To>(std::declval<From>()))>>
  : std::true_type {};

/// Low `bits` of `wide`, saturated to all-ones when `wide` does not fit.
template <typename Raw, std::size_t Bits>
Raw saturate_low_bits(uint256_t wide) noexcept
{
    static_assert(Bits > 0 && Bits <= 256);
    if constexpr (Bits <= 64)
    {
        const std::uint64_t low = wide.lower().lower();
        const bool overflow = wide.upper() != uint128_t{0}
            || wide.lower().upper() != 0
            || (Bits < 64 && (low >> Bits) != 0);
        const std::uint64_t mag = overflow
            ? (Bits == 64 ? ~std::uint64_t{0} : ((std::uint64_t{1} << Bits) - 1))
            : low;
        return static_cast<Raw>(mag);
    }
    else if constexpr (Bits <= 128)
    {
        const std::uint64_t lo = wide.lower().lower();
        const std::uint64_t hi = wide.lower().upper();
        const bool overflow = wide.upper() != uint128_t{0}
            || (Bits < 128 && (hi >> (Bits - 64)) != 0);
        simde_uint128 mag;
        if (overflow)
        {
            if constexpr (Bits == 128)
                mag = ~simde_uint128{0};
            else
                mag = (simde_uint128{1} << Bits) - 1;
        }
        else
        {
            mag = (static_cast<simde_uint128>(hi) << 64) | lo;
        }
        return static_cast<Raw>(mag);
    }
    else
    {
        uint256_t mag = wide;
        if constexpr (Bits < 256)
        {
            const auto sh = static_cast<unsigned>(Bits - 128);
            const bool overflow = (wide.upper() >> sh) != uint128_t{0};
            if (overflow)
            {
                const uint128_t hi = (uint128_t{1} << sh) - 1;
                mag = uint256_t{hi, ~uint128_t{0}};
            }
        }
        return static_cast<Raw>(mag);
    }
}

template <typename IntegralType>
inline IntegralType rounded_double_to_integral(double rounded) noexcept
{
    if constexpr (std::is_integral_v<IntegralType>
        || std::is_same_v<IntegralType, simde_int128>
        || std::is_same_v<IntegralType, simde_uint128>)
    {
        return static_cast<IntegralType>(rounded);
    }
    else
    {
        const bool neg = std::signbit(rounded);
        const double mag = neg ? -rounded : rounded;
        const uint256_t wide = uint256_from_rounded_double(mag);
        IntegralType result{};
        if constexpr (std::is_same_v<IntegralType, uint128_t>)
        {
            result = wide.upper() != uint128_t{0} ? ~uint128_t{0} : wide.lower();
        }
        else if constexpr (std::is_same_v<IntegralType, uint256_t>)
        {
            result = wide;
        }
        else
        {
            constexpr std::size_t bits = dpf::utils::bitlength_of_v<IntegralType>;
            using raw_type = typename dpf::utils::make_from_integral_value<IntegralType>::integral_type;
            if constexpr (bits > 0 && bits <= 256
                && is_static_castable<IntegralType, raw_type>::value)
            {
                const raw_type raw = saturate_low_bits<raw_type, bits>(wide);
                result = dpf::utils::make_from_integral_value<IntegralType>{}(raw);
            }
            else
            {
                static_assert(bits > 0 && bits <= 256
                    && is_static_castable<IntegralType, raw_type>::value,
                    "fixedpoint construction from double is not implemented for this integer type");
                return IntegralType{};
            }
        }
        if (neg)
        {
            result = -result;
        }
        return result;
    }
}

/// Shift an integer into fixed-point raw form: `value * 2^FractionalBits`,
/// wrapping in the backend's two's-complement encoding. One shift; no `double`.
template <typename IntegralType,
          unsigned FractionalBits,
          typename T>
HEDLEY_ALWAYS_INLINE
HEDLEY_CONST
constexpr IntegralType scale_integer_to_fixed_raw(T integer_value) noexcept
{
    using unsigned_type = dpf::utils::make_unsigned_t<IntegralType>;
    const auto bits = static_cast<unsigned_type>(
        static_cast<IntegralType>(integer_value));
    if constexpr (FractionalBits == 0)
    {
        return static_cast<IntegralType>(bits);
    }
    return static_cast<IntegralType>(bits << FractionalBits);
}

template <typename IntegralType>
inline constexpr bool is_signed_rep_v =
    std::is_signed_v<IntegralType>
    || std::is_same_v<IntegralType, simde_int128>;

/// Two's-complement negate via the unsigned width. Defined for the
/// most-negative value (wraps); signed `-x` would be UB there.
template <typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_CONST
constexpr IntegralType raw_neg(IntegralType x) noexcept
{
    using unsigned_type = dpf::utils::make_unsigned_t<IntegralType>;
    return static_cast<IntegralType>(
        unsigned_type{} - static_cast<unsigned_type>(x));
}

template <typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_CONST
constexpr IntegralType raw_abs(IntegralType x) noexcept
{
    if constexpr (is_signed_rep_v<IntegralType>)
    {
        return (x < IntegralType{}) ? raw_neg(x) : x;
    }
    return x;
}

/// Remainder with the sign of `a` and magnitude `< |b|` (C++ `%` /
/// `std::fmod`). Zero divisor → 0; this type has no NaN.
template <typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_CONST
constexpr IntegralType raw_fmod(IntegralType a, IntegralType b) noexcept
{
    if (b == IntegralType{})
    {
        return IntegralType{};
    }
    return a % b;
}

}  // namespace detail

template <unsigned FractionalBits,
          typename IntegralType>
auto constexpr make_fixed_from_integral_type(IntegralType value) noexcept;

/// @tparam FractionalBits Number of fractional bits used in the fixed-point
///         representation.
/// @tparam IntegralType The underlying integral type used for the fixed-point
///         representation.
template <unsigned FractionalBits,
          typename IntegralType = GROTTO_FIXED_DEFAULT_INTEGRAL_REPRESENTATION>
struct fixedpoint
{
    using integral_type = IntegralType;
    static constexpr int fractional_bits = FractionalBits;
    static constexpr int integer_bits = dpf::utils::bitlength_of_v<integral_type> - fractional_bits;

    static_assert(std::numeric_limits<integral_type>::is_integer
        || std::is_same_v<integral_type, simde_int128>
        || std::is_same_v<integral_type, simde_uint128>);
    static_assert(fractional_bits <= dpf::utils::bitlength_of_v<integral_type>);
    static_assert(integer_bits >= 0);

private:
    struct raw_tag {};

public:

    /// @name C'tors
    /// @brief Constructs a new fixed-point number.
    /// @{

    /// @brief Default c'tor
    /// @details Zero-initializes the encoding (`fixedpoint x;` is 0).
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint() noexcept = default;

    /// @brief Copy c'tor
    /// @details Constructs a fixed-point with the value copied from `other`.
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint(const fixedpoint & other) noexcept = default;

    /// @brief Move c'tor
    /// @details Constructs a fixed-point with the value copied from `other` using move semantics.
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint(fixedpoint && other) noexcept = default;

    /// @brief Value c'tor
    /// @details Initializes the fixed-point with the value determined by `desired`, using the <a href="https://en.cppreference.com/w/cpp/numeric/fenv/FE_round">current rounding mode</a> for the least-significant bit.
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint(double desired) noexcept  // NOLINT (implicit c'tor)
      : value{detail::rounded_double_to_integral<integral_type>(
            std::nearbyint(std::ldexp(desired, fractional_bits)))}
    { }

    /// @brief Integer value c'tor
    /// @details `fixedpoint(3)` is the mathematical value 3 (raw encoding
    ///          `3 << fractional_bits`), not a raw word. One shift; no `double`.
    ///          Use `from_raw` for a bit-exact encoding.
    template <typename T,
              std::enable_if_t<
                  std::is_integral_v<T>
                  && !std::is_same_v<std::remove_cv_t<T>, bool>, int> = 0>
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint(T integer_value) noexcept  // NOLINT (implicit c'tor)
      : value{detail::scale_integer_to_fixed_raw<
            integral_type, static_cast<unsigned>(fractional_bits)>(integer_value)}
    { }

    /// @brief Bit-exact construction from the backend integer encoding.
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    static constexpr fixedpoint from_raw(integral_type raw) noexcept
    {
        return fixedpoint{raw, raw_tag{}};
    }

    /// @}

    /// @name Assignment operators
    /// @brief Assign a new value to a fixed-point number
    /// {@

    /// @brief Copy assignment
    /// @details Assigns the fixed-point with a copy of `other`
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint & operator=(const fixedpoint & other) noexcept = default;

    /// @brief Move assignment
    /// @details Assigns the fixed-point with a copy of `other` using move semantics.
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint & operator=(fixedpoint && other) noexcept = default;

    /// @brief Value assignment
    /// @details Assigns the fixed-point with a value determined by `desired`, using the <a href="https://en.cppreference.com/w/cpp/numeric/fenv/FE_round">current rounding mode</a> for the least-significant bit..
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint & operator=(const double & desired) noexcept
    {
        value = detail::rounded_double_to_integral<integral_type>(
            std::nearbyint(std::ldexp(desired, fractional_bits)));
        return *this;
    }

    /// @}

    ~fixedpoint() = default;

    /// @brief Cast to `double`
    HEDLEY_ALWAYS_INLINE
    HEDLEY_PURE
    explicit constexpr operator double() const noexcept
    {
        return std::ldexp(static_cast<double>(value), -static_cast<double>(fractional_bits));
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator&(integral_type mask) const noexcept
    {
        return static_cast<bool>(this->integral_representation() & mask);
    }

    /// Bit test against another encoding (DPF writes `mask & x` with both
    /// sides the input type when `msb_mask` is a `fixedpoint`).
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator&(fixedpoint mask) const noexcept
    {
        return static_cast<bool>(value & mask.value);
    }

    /// Bitwise complement of the encoding. `std::bit_not` uses this.
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr fixedpoint operator~() const noexcept
    {
        using unsigned_type = dpf::utils::make_unsigned_t<integral_type>;
        return from_raw(static_cast<integral_type>(
            ~static_cast<unsigned_type>(value)));
    }

    /// Next / previous representable encoding (one ULP).
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint & operator++() noexcept
    {
        ++value;
        return *this;
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint operator++(int) noexcept
    {
        fixedpoint tmp = *this;
        ++*this;
        return tmp;
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint & operator--() noexcept
    {
        --value;
        return *this;
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint operator--(int) noexcept
    {
        fixedpoint tmp = *this;
        --*this;
        return tmp;
    }

    /// Logical shift of the encoding. DPF walks `msb_mask` with `>>`; a
    /// signed arithmetic shift would sign-extend the MSB and break that.
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint & operator>>=(std::size_t n) noexcept
    {
        using unsigned_type = dpf::utils::make_unsigned_t<integral_type>;
        value = static_cast<integral_type>(static_cast<unsigned_type>(value) >> n);
        return *this;
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint & operator<<=(std::size_t n) noexcept
    {
        using unsigned_type = dpf::utils::make_unsigned_t<integral_type>;
        value = static_cast<integral_type>(static_cast<unsigned_type>(value) << n);
        return *this;
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    friend constexpr fixedpoint operator>>(fixedpoint x, std::size_t n) noexcept
    {
        x >>= n;
        return x;
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    friend constexpr fixedpoint operator<<(fixedpoint x, std::size_t n) noexcept
    {
        x <<= n;
        return x;
    }

    /// @brief Access underlying integral representation
    /// @details If the represented fixed-point number is `x`, then this
    ///          function returns an `integral_type` whose value is `x*2**fractional_bits`.
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr integral_type integral_representation() const noexcept
    {
        return this->value;
    }

    /// @brief Unary negation operator
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr fixedpoint operator-() const noexcept
    {
        return from_raw(detail::raw_neg(value));
    }

    /// @brief Binary addition operator
    /// @details Computes the sum of two fixed-point numbers
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr fixedpoint operator+(fixedpoint rhs) const noexcept
    {
        return from_raw(value + rhs.value);
    }

    /// @brief Binary addition assignment operator
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint & operator+=(fixedpoint rhs) noexcept
    {
        this->value += rhs.integral_representation();
        return *this;
    }

    /// @brief Binary subtraction operator
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr fixedpoint operator-(fixedpoint rhs) const noexcept
    {
        return from_raw(value - rhs.value);
    }

    /// @brief Binary addition assignment operator
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint & operator-=(fixedpoint rhs) noexcept
    {
        this->value -= rhs.integral_representation();
        return *this;
    }

    /// @brief Binary multiplication operator
    template <unsigned FractionalBits1>
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr auto operator*(fixedpoint<FractionalBits1, IntegralType> rhs) const noexcept
    {
        return make_fixed_from_integral_type<FractionalBits + FractionalBits1>((this->integral_representation() * rhs.integral_representation()));
    }

    /// @name Equality
    /// @brief Strict equality operator
    /// @{
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator==(fixedpoint rhs) const noexcept
    {
        return (this->integral_representation() == rhs.integral_representation());
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator==(double rhs) const noexcept
    {
        return is_in_range(rhs) && (*this == fixedpoint(rhs));
    }
    /// @}

    /// @name Inequality
    /// @brief Strict inequality operator
    /// @{
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator!=(fixedpoint rhs) const noexcept
    {
        return (this->integral_representation() != rhs.integral_representation());
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator!=(double rhs) const noexcept
    {
        return !(*this == rhs);
    }
    /// @}

    /// @name Less than
    /// @brief Binary less-than operator
    /// @{
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator<(fixedpoint rhs) const noexcept
    {
        return (this->integral_representation() < rhs.integral_representation());
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator<(double rhs) const noexcept
    {
        if (std::isnan(rhs))
            return false;
        if (!is_in_range(rhs))
            return rhs > static_cast<double>(std::numeric_limits<fixedpoint>::max());
        return *this < fixedpoint(rhs);
    }
    /// @}

    /// @name Less than or equal
    /// @brief Binary less-than-or-equal operator
    /// @{
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator<=(fixedpoint rhs) const noexcept
    {
        return (this->integral_representation() <= rhs.integral_representation());
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator<=(double rhs) const noexcept
    {
        if (std::isnan(rhs))
            return false;
        if (!is_in_range(rhs))
            return rhs > static_cast<double>(std::numeric_limits<fixedpoint>::max());
        return *this <= fixedpoint(rhs);
    }
    /// @}

    /// @name Greater than
    /// @brief Binary greater-than operator
    /// @{
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator>(fixedpoint rhs) const noexcept
    {
        return (this->integral_representation() > rhs.integral_representation());
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator>(double rhs) const noexcept
    {
        if (std::isnan(rhs))
            return false;
        if (!is_in_range(rhs))
            return rhs < static_cast<double>(std::numeric_limits<fixedpoint>::lowest());
        return *this > fixedpoint(rhs);
    }
    /// @}

    /// @name Greater than or equal
    /// @brief Binary greater-than-or-equal operator
    /// @{
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator>=(fixedpoint rhs) const noexcept
    {
        return (this->integral_representation() >= rhs.integral_representation());
    }

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_PURE
    constexpr bool operator>=(double rhs) const noexcept
    {
        if (std::isnan(rhs))
            return false;
        if (!is_in_range(rhs))
            return rhs < static_cast<double>(std::numeric_limits<fixedpoint>::lowest());
        return *this >= fixedpoint(rhs);
    }
    /// @}

  private:
    // struct make_fixed_from_integral_type_tag {};

    /// @brief Determine if a floating-point is within range
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    static constexpr bool is_in_range(double d) noexcept
    {
        return HEDLEY_LIKELY(
            static_cast<double>(std::numeric_limits<fixedpoint>::lowest()) <= d
              && d <= static_cast<double>(std::numeric_limits<fixedpoint>::max()));
    }

    // HEDLEY_ALWAYS_INLINE
    // HEDLEY_NO_THROW
    // constexpr explicit fixedpoint(integral_type val, make_fixed_from_integral_type_tag &&) noexcept
    //   : value{val}
    // { }

    // template <unsigned F, typename T> friend constexpr auto nextafter(fixedpoint<F, T>) noexcept;
    // template <unsigned F, typename T> friend constexpr auto nextbefore(fixedpoint<F, T>) noexcept;
    // template <unsigned F0, unsigned F1, typename T> friend constexpr auto precision_cast(fixedpoint<F1, T>) noexcept;
    template <unsigned F, typename T> friend auto constexpr make_fixed_from_integral_type(T) noexcept;
    // template <unsigned F, typename T> friend constexpr auto fabs(fixedpoint<F, T>) noexcept;
    // template <unsigned F, typename T> friend constexpr auto fmod(fixedpoint<F, T>, double) noexcept;

    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    constexpr fixedpoint(integral_type raw, raw_tag) noexcept
      : value{raw}
    { }

    integral_type value{};
};

/// @brief Bit test with the mask on the left. DPF key generation and
///        evaluation write `mask & x`.
template <unsigned FractionalBits,
          typename IntegralType,
          typename Mask>
constexpr bool operator&(const Mask & mask,
    const fixedpoint<FractionalBits, IntegralType> & x) noexcept
{
    return static_cast<bool>(x.integral_representation()
        & static_cast<IntegralType>(mask));
}

template <class CharT,
          class Traits,
          unsigned FractionalBits,
          typename IntegralType>
std::basic_ostream<CharT, Traits> &
operator<<(std::basic_ostream<CharT, Traits> & os,
    const fixedpoint<FractionalBits, IntegralType> & f) noexcept
{
    return os << static_cast<double>(f);
}

template <class CharT,
          class Traits,
          unsigned FractionalBits,
          typename IntegralType>
std::basic_istream<CharT, Traits> &
operator>>(std::basic_istream<CharT, Traits> & is,
    fixedpoint<FractionalBits, IntegralType> & f)
{
    double d;
    is >> d;
    f = d;
    return is;
}

template <unsigned FractionalBits,
          typename IntegralType>
auto constexpr make_fixed_from_integral_type(IntegralType value) noexcept
{
    return fixedpoint<FractionalBits, IntegralType>::from_raw(value);
}

template <unsigned FractionalBits,
          typename IntegralType = GROTTO_FIXED_DEFAULT_INTEGRAL_REPRESENTATION>
HEDLEY_ALWAYS_INLINE
HEDLEY_CONST
static constexpr auto make_fixed(double d)
{
    return fixedpoint<FractionalBits, IntegralType>(d);
}

template <typename FixedType>
HEDLEY_ALWAYS_INLINE
HEDLEY_CONST
static constexpr auto make_fixed(double d)
{
    return fixedpoint<FixedType::fractional_bits, typename FixedType::integral_type>(d);
}

/// @brief Creates a fixed-point number from a double with bounds checking.
/// @throws std::range_error If the input double is outside the representable
///         range of the fixed-point number.
template <unsigned FractionalBits,
          typename IntegralType = GROTTO_FIXED_DEFAULT_INTEGRAL_REPRESENTATION>
HEDLEY_ALWAYS_INLINE
HEDLEY_CONST
static auto make_fixed_safe(double d)
{
    using fixed_type = fixedpoint<FractionalBits, IntegralType>;

    if (HEDLEY_UNLIKELY(d < std::numeric_limits<fixed_type>::lowest()))
    {
        throw std::range_error("value is too small (underflows integral representation)");
    }

    if (HEDLEY_UNLIKELY(std::numeric_limits<fixed_type>::max() < d))
    {
        throw std::range_error("value is too large (overflows integral representation)");
    }

    return make_fixed<FractionalBits, IntegralType>(d);
}

template <unsigned ToFractionalBits,
          unsigned FromFractionalBits,
          typename IntegralType>
constexpr auto precision_cast(const fixedpoint<FromFractionalBits, IntegralType> & f) noexcept
{
    auto value = f.integral_representation();
    if constexpr (ToFractionalBits > FromFractionalBits)
    {
        return make_fixed_from_integral_type<ToFractionalBits, IntegralType>(value << (ToFractionalBits - FromFractionalBits));
    }
    else
    {
        return make_fixed_from_integral_type<ToFractionalBits, IntegralType>(value >> (FromFractionalBits - ToFractionalBits));
    }
}

template <unsigned FractionalBits,
          typename IntegralType>
static constexpr auto precision_of(fixedpoint<FractionalBits, IntegralType>) noexcept
{
    return FractionalBits;
}

template <unsigned FractionalBits,
          typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_NO_THROW
HEDLEY_CONST
constexpr auto nextafter(fixedpoint<FractionalBits, IntegralType> f) noexcept
{
    return make_fixed_from_integral_type<FractionalBits, IntegralType>(f.integral_representation()+1);
}

template <unsigned FractionalBits,
          typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_NO_THROW
HEDLEY_CONST
constexpr auto nextbefore(fixedpoint<FractionalBits, IntegralType> f) noexcept
{
    return make_fixed_from_integral_type<FractionalBits, IntegralType>(f.integral_representation()-1);
}

template <unsigned FractionalBits,
          typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_NO_THROW
HEDLEY_CONST
constexpr auto fabs(fixedpoint<FractionalBits, IntegralType> v) noexcept
{
    if constexpr (!detail::is_signed_rep_v<IntegralType>)
    {
        return v;
    }
    const auto mag = v.integral_representation();
    if (mag < IntegralType{})
    {
        return fixedpoint<FractionalBits, IntegralType>::from_raw(
            detail::raw_neg(mag));
    }
    return v;
}

template <unsigned FractionalBits,
          typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_NO_THROW
HEDLEY_CONST
constexpr auto fmod(fixedpoint<FractionalBits, IntegralType> v,
    fixedpoint<FractionalBits, IntegralType> modulus) noexcept
{
    return fixedpoint<FractionalBits, IntegralType>::from_raw(
        detail::raw_fmod(v.integral_representation(),
            modulus.integral_representation()));
}

template <unsigned FractionalBits,
          typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_NO_THROW
HEDLEY_CONST
constexpr auto fmod(fixedpoint<FractionalBits, IntegralType> v,
    double modulus) noexcept
{
    return fmod(v, make_fixed<FractionalBits, IntegralType>(modulus));
}

template <unsigned FractionalBits,
          typename IntegralType,
          typename T,
          std::enable_if_t<
              std::is_integral_v<T>
              && !std::is_same_v<std::remove_cv_t<T>, bool>, int> = 0>
HEDLEY_ALWAYS_INLINE
HEDLEY_NO_THROW
HEDLEY_CONST
constexpr auto fmod(fixedpoint<FractionalBits, IntegralType> v,
    T modulus) noexcept
{
    return fmod(v, fixedpoint<FractionalBits, IntegralType>(modulus));
}

enum fixed_cast_policy
{
    use_default,
    use_left_arg,
    use_right_arg,
    use_min_arg,
    use_max_arg,
    use_arg_sum  //< for multiplies only
};

template <typename BinaryOperator,
          fixed_cast_policy Mode = use_max_arg>
struct binary_operator_precast_wrapper
{
    static_assert(Mode == use_default   ||
                  Mode == use_left_arg  ||
                  Mode == use_right_arg ||
                  Mode == use_min_arg   ||
                  Mode == use_max_arg);

    template <typename IntegralType,
              unsigned FractionalBitsLHS,
              unsigned FractionalBitsRHS>
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_CONST
    constexpr auto operator()(
        fixedpoint<FractionalBitsLHS, IntegralType> lhs,
        fixedpoint<FractionalBitsRHS, IntegralType> rhs) const noexcept
    {
        constexpr bool lt = (FractionalBitsLHS < FractionalBitsRHS);
        constexpr bool eq = (FractionalBitsLHS == FractionalBitsRHS);
        constexpr BinaryOperator op{};

        if constexpr (eq) return op(lhs, rhs);      // no cast necessary
        else if constexpr ((Mode == use_left_arg)   // always casting to lhs
            || ((Mode == use_min_arg) &&  lt)       // lhs happens to be min
            || ((Mode == use_max_arg) && !lt)       // lhs happens to be max
            || ((Mode == use_default) && !lt))      // default == use_max
        {
            return op(lhs, precision_cast<FractionalBitsLHS>(rhs));
        }
        else if constexpr ((Mode == use_right_arg)  // always casting to rhs
            || ((Mode == use_min_arg) && !lt)       // rhs happens to be min
            || ((Mode == use_max_arg) &&  lt)       // rhs happens to be max
            || ((Mode == use_default) &&  lt))      // default == use_max
        {
            return op(precision_cast<FractionalBitsRHS>(lhs), rhs);
        }
        else
        {
            HEDLEY_UNREACHABLE();
        }
    }
};

template <fixed_cast_policy Mode = use_arg_sum>
struct multiplies
{
    static_assert(Mode == use_default   ||
                  Mode == use_left_arg  ||
                  Mode == use_right_arg ||
                  Mode == use_min_arg   ||
                  Mode == use_max_arg   ||
                  Mode == use_arg_sum);

    template <unsigned FractionalBitsOut = unsigned(-1),
              typename IntegralType,
              unsigned FractionalBitsLHS,
              unsigned FractionalBitsRHS>
    HEDLEY_ALWAYS_INLINE
    HEDLEY_NO_THROW
    HEDLEY_CONST
    constexpr auto operator()(
        fixedpoint<FractionalBitsLHS, IntegralType> lhs,
        fixedpoint<FractionalBitsRHS, IntegralType> rhs) const noexcept
    {
        // First element is the right-shift that brings the raw product down to
        // `FractionalBitsOut`. It is negative when the product must be shifted left.
        constexpr auto r = []() -> std::pair<int, unsigned>
        {
            constexpr bool lt = (FractionalBitsLHS < FractionalBitsRHS);
            constexpr bool eq = (FractionalBitsLHS == FractionalBitsRHS);
            if constexpr (FractionalBitsOut != unsigned(-1))
            {
                return {static_cast<int>(FractionalBitsLHS + FractionalBitsRHS)
                        - static_cast<int>(FractionalBitsOut),
                    FractionalBitsOut};
            }
            if constexpr (eq)
            {
                return {static_cast<int>((FractionalBitsLHS+FractionalBitsLHS)/2),
                    1u+(FractionalBitsLHS+FractionalBitsLHS-1)/2};
            }
            if constexpr ((Mode == use_left_arg)
                || ((Mode == use_min_arg) &&  lt)
                || ((Mode == use_max_arg) && !lt))
            {
                return {static_cast<int>(FractionalBitsRHS), FractionalBitsLHS};
            }
            if constexpr ((Mode == use_right_arg)
                || ((Mode == use_min_arg) && !lt)
                || ((Mode == use_max_arg) &&  lt))
            {
                return {static_cast<int>(FractionalBitsLHS), FractionalBitsRHS};
            }
            else if constexpr ((Mode == use_arg_sum) || (Mode == use_default))
            {
                return {0, FractionalBitsRHS+FractionalBitsLHS};
            }
        }();

        if constexpr (std::numeric_limits<IntegralType>::digits > 32)
        {
            simde_uint128 product = simde_uint128(lhs.integral_representation())
                * rhs.integral_representation();
            if constexpr (r.first < 0)
                product = product << -r.first;
            else
                product = product >> r.first;
            return make_fixed_from_integral_type<r.second, IntegralType>(
                static_cast<IntegralType>(product));
        }
        else
        {
            std::uint64_t product = std::uint64_t(lhs.integral_representation())
                * rhs.integral_representation();
            if constexpr (r.first < 0)
                product = product << -r.first;
            else
                product = product >> static_cast<unsigned>(r.first);
            return make_fixed_from_integral_type<r.second, IntegralType>(
                static_cast<IntegralType>(product));
        }
    }
};

template <unsigned FractionalBits,
          typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_NO_THROW
HEDLEY_CONST
constexpr bool operator==(double lhs, fixedpoint<FractionalBits, IntegralType> rhs) noexcept
{
    return (rhs == lhs);
}

template <unsigned FractionalBits,
          typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_NO_THROW
HEDLEY_CONST
constexpr bool operator<(double lhs, fixedpoint<FractionalBits, IntegralType> rhs) noexcept
{
    return (rhs > lhs);
}

template <unsigned FractionalBits,
          typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_NO_THROW
HEDLEY_CONST
constexpr bool operator<=(double lhs, fixedpoint<FractionalBits, IntegralType> rhs) noexcept
{
    return (rhs >= lhs);
}

template <unsigned FractionalBits,
          typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_NO_THROW
HEDLEY_CONST
constexpr bool operator>(double lhs, fixedpoint<FractionalBits, IntegralType> rhs) noexcept
{
    return (rhs < lhs);
}

template <unsigned FractionalBits,
          typename IntegralType>
HEDLEY_ALWAYS_INLINE
HEDLEY_NO_THROW
HEDLEY_CONST
constexpr bool operator>=(double lhs, fixedpoint<FractionalBits, IntegralType> rhs) noexcept
{
    return (rhs <= lhs);
}

template <typename FixedPointType,
          std::size_t Degree>
struct fixedpoint_polynomial : public std::array<FixedPointType, Degree>
{
    using coefficient_type = FixedPointType;
    static constexpr std::size_t degree = Degree;
    constexpr auto operator()(coefficient_type x) const
    {
        constexpr auto product_of = multiplies<use_arg_sum>{};
        constexpr auto sum_of = binary_operator_precast_wrapper<std::plus<coefficient_type>, use_max_arg>{};
        auto coeff = this->rbegin();
        coefficient_type y{*coeff};
        while (++coeff != this->rend())
        {
            y = sum_of(product_of(y, x), *coeff);
        }
        return y;
    }
};

template <typename FixedPointType,
          std::size_t Degree>
static constexpr auto evaluate(const fixedpoint_polynomial<FixedPointType, Degree> & poly, FixedPointType x)
{
    return poly(x);
}

namespace fixedpoint_literals
{

constexpr auto operator "" _fixed0(long double val)
{
    return grotto::make_fixed<0>(val);
}

constexpr auto operator "" _fixed1(long double val)
{
    return grotto::make_fixed<1>(val);
}

constexpr auto operator "" _fixed2(long double val)
{
    return grotto::make_fixed<2>(val);
}

constexpr auto operator "" _fixed3(long double val)
{
    return grotto::make_fixed<3>(val);
}

constexpr auto operator "" _fixed4(long double val)
{
    return grotto::make_fixed<4>(val);
}

constexpr auto operator "" _fixed5(long double val)
{
    return grotto::make_fixed<5>(val);
}

constexpr auto operator "" _fixed6(long double val)
{
    return grotto::make_fixed<6>(val);
}

constexpr auto operator "" _fixed7(long double val)
{
    return grotto::make_fixed<7>(val);
}

constexpr auto operator "" _fixed8(long double val)
{
    return grotto::make_fixed<8>(val);
}

constexpr auto operator "" _fixed9(long double val)
{
    return grotto::make_fixed<9>(val);
}

constexpr auto operator "" _fixed10(long double val)
{
    return grotto::make_fixed<10>(val);
}

constexpr auto operator "" _fixed11(long double val)
{
    return grotto::make_fixed<11>(val);
}

constexpr auto operator "" _fixed12(long double val)
{
    return grotto::make_fixed<12>(val);
}

constexpr auto operator "" _fixed13(long double val)
{
    return grotto::make_fixed<13>(val);
}

constexpr auto operator "" _fixed14(long double val)
{
    return grotto::make_fixed<14>(val);
}

constexpr auto operator "" _fixed15(long double val)
{
    return grotto::make_fixed<15>(val);
}

constexpr auto operator "" _fixed16(long double val)
{
    return grotto::make_fixed<16>(val);
}

constexpr auto operator "" _fixed17(long double val)
{
    return grotto::make_fixed<17>(val);
}

constexpr auto operator "" _fixed18(long double val)
{
    return grotto::make_fixed<18>(val);
}

constexpr auto operator "" _fixed19(long double val)
{
    return grotto::make_fixed<19>(val);
}

constexpr auto operator "" _fixed20(long double val)
{
    return grotto::make_fixed<20>(val);
}

constexpr auto operator "" _fixed21(long double val)
{
    return grotto::make_fixed<21>(val);
}

constexpr auto operator "" _fixed22(long double val)
{
    return grotto::make_fixed<22>(val);
}

constexpr auto operator "" _fixed23(long double val)
{
    return grotto::make_fixed<23>(val);
}

constexpr auto operator "" _fixed24(long double val)
{
    return grotto::make_fixed<24>(val);
}

constexpr auto operator "" _fixed25(long double val)
{
    return grotto::make_fixed<25>(val);
}

constexpr auto operator "" _fixed26(long double val)
{
    return grotto::make_fixed<26>(val);
}

constexpr auto operator "" _fixed27(long double val)
{
    return grotto::make_fixed<27>(val);
}

constexpr auto operator "" _fixed28(long double val)
{
    return grotto::make_fixed<28>(val);
}

constexpr auto operator "" _fixed29(long double val)
{
    return grotto::make_fixed<29>(val);
}

constexpr auto operator "" _fixed30(long double val)
{
    return grotto::make_fixed<30>(val);
}

constexpr auto operator "" _fixed31(long double val)
{
    return grotto::make_fixed<31>(val);
}

constexpr auto operator "" _fixed32(long double val)
{
    return grotto::make_fixed<32>(val);
}

constexpr auto operator "" _fixed33(long double val)
{
    return grotto::make_fixed<33>(val);
}
constexpr auto operator "" _fixed34(long double val)
{
    return grotto::make_fixed<34>(val);
}
constexpr auto operator "" _fixed35(long double val)
{
    return grotto::make_fixed<35>(val);
}
constexpr auto operator "" _fixed36(long double val)
{
    return grotto::make_fixed<36>(val);
}
constexpr auto operator "" _fixed37(long double val)
{
    return grotto::make_fixed<37>(val);
}
constexpr auto operator "" _fixed38(long double val)
{
    return grotto::make_fixed<38>(val);
}
constexpr auto operator "" _fixed39(long double val)
{
    return grotto::make_fixed<39>(val);
}
constexpr auto operator "" _fixed40(long double val)
{
    return grotto::make_fixed<40>(val);
}
constexpr auto operator "" _fixed41(long double val)
{
    return grotto::make_fixed<41>(val);
}
constexpr auto operator "" _fixed42(long double val)
{
    return grotto::make_fixed<42>(val);
}
constexpr auto operator "" _fixed43(long double val)
{
    return grotto::make_fixed<43>(val);
}
constexpr auto operator "" _fixed44(long double val)
{
    return grotto::make_fixed<44>(val);
}
constexpr auto operator "" _fixed45(long double val)
{
    return grotto::make_fixed<45>(val);
}
constexpr auto operator "" _fixed46(long double val)
{
    return grotto::make_fixed<46>(val);
}
constexpr auto operator "" _fixed47(long double val)
{
    return grotto::make_fixed<47>(val);
}
constexpr auto operator "" _fixed48(long double val)
{
    return grotto::make_fixed<48>(val);
}
constexpr auto operator "" _fixed49(long double val)
{
    return grotto::make_fixed<49>(val);
}
constexpr auto operator "" _fixed50(long double val)
{
    return grotto::make_fixed<50>(val);
}
constexpr auto operator "" _fixed51(long double val)
{
    return grotto::make_fixed<51>(val);
}
constexpr auto operator "" _fixed52(long double val)
{
    return grotto::make_fixed<52>(val);
}
constexpr auto operator "" _fixed53(long double val)
{
    return grotto::make_fixed<53>(val);
}
constexpr auto operator "" _fixed54(long double val)
{
    return grotto::make_fixed<54>(val);
}
constexpr auto operator "" _fixed55(long double val)
{
    return grotto::make_fixed<55>(val);
}
constexpr auto operator "" _fixed56(long double val)
{
    return grotto::make_fixed<56>(val);
}
constexpr auto operator "" _fixed57(long double val)
{
    return grotto::make_fixed<57>(val);
}
constexpr auto operator "" _fixed58(long double val)
{
    return grotto::make_fixed<58>(val);
}
constexpr auto operator "" _fixed59(long double val)
{
    return grotto::make_fixed<59>(val);
}
constexpr auto operator "" _fixed60(long double val)
{
    return grotto::make_fixed<60>(val);
}
constexpr auto operator "" _fixed61(long double val)
{
    return grotto::make_fixed<61>(val);
}
constexpr auto operator "" _fixed62(long double val)
{
    return grotto::make_fixed<62>(val);
}
constexpr auto operator "" _fixed63(long double val)
{
    return grotto::make_fixed<63>(val);
}
constexpr auto operator "" _fixed64(long double val)
{
    return grotto::make_fixed<64>(val);
}

}  // namespace grotto::fixedpoint_literals

}  // namespace grotto

using grotto::precision_cast;

namespace dpf
{

namespace utils
{

template <unsigned FractionalBits,
          typename IntegralType>
struct bitlength_of<grotto::fixedpoint<FractionalBits, IntegralType>>
  : public bitlength_of<IntegralType> { };

template <unsigned FractionalBits,
          typename IntegralType>
struct msb_of<grotto::fixedpoint<FractionalBits, IntegralType>>
{
    static constexpr grotto::fixedpoint<FractionalBits, IntegralType> value =
        grotto::fixedpoint<FractionalBits, IntegralType>::from_raw(
            static_cast<IntegralType>(msb_of_v<IntegralType>));
};

template <unsigned FractionalBits,
          typename IntegralType>
struct uses_signed_msb<grotto::fixedpoint<FractionalBits, IntegralType>>
  : std::bool_constant<uses_signed_msb_v<IntegralType>> { };

template <unsigned FractionalBits,
          typename IntegralType>
struct make_unsigned<grotto::fixedpoint<FractionalBits, IntegralType>>
{
    using type = grotto::fixedpoint<FractionalBits, make_unsigned_t<IntegralType>>;
};

template <unsigned FractionalBits,
          typename IntegralType>
struct countl_zero_symmetric_difference<grotto::fixedpoint<FractionalBits, IntegralType>>
{
    using T = grotto::fixedpoint<FractionalBits, IntegralType>;
    static constexpr auto clz = dpf::utils::countl_zero_symmetric_difference<typename T::integral_type>{};

    HEDLEY_PURE
    HEDLEY_ALWAYS_INLINE
    constexpr std::size_t operator()(const T & lhs, const T & rhs) const noexcept
    {
        return clz(lhs.integral_representation(), rhs.integral_representation());
    }
};

template <unsigned FractionalBits,
          typename IntegralType>
struct to_integral_type<grotto::fixedpoint<FractionalBits, IntegralType>>
    : to_integral_type_base<grotto::fixedpoint<FractionalBits, IntegralType>>
{
    using parent = to_integral_type_base<grotto::fixedpoint<FractionalBits, IntegralType>>;
    using typename parent::integral_type;

    HEDLEY_PURE
    HEDLEY_NO_THROW
    HEDLEY_ALWAYS_INLINE
    constexpr integral_type operator()(
        const grotto::fixedpoint<FractionalBits, IntegralType> & input) const noexcept
    {
        return to_integral_type<IntegralType>{}(input.integral_representation());
    }
};

template <unsigned FractionalBits,
          typename IntegralType>
struct mod_pow_2<grotto::fixedpoint<FractionalBits, IntegralType>>
{
    using fixed_type = grotto::fixedpoint<FractionalBits, IntegralType>;
    static constexpr auto mod = mod_pow_2<IntegralType>{};

    std::size_t operator()(fixed_type val, std::size_t n) const noexcept
    {
        return mod(val.integral_representation(), n);
    }
};

template <unsigned FractionalBits,
          typename IntegralType>
struct make_from_integral_value<grotto::fixedpoint<FractionalBits, IntegralType>>
{
    using fixed_type = grotto::fixedpoint<FractionalBits, IntegralType>;
    using integral_type = typename to_integral_type<fixed_type>::integral_type;

    constexpr fixed_type operator()(integral_type val) const noexcept
    {
        return grotto::make_fixed_from_integral_type<FractionalBits, IntegralType>(
            static_cast<IntegralType>(val));
    }
};

template <unsigned FractionalBits,
          typename IntegralType>
struct flip_msb_for_input<grotto::fixedpoint<FractionalBits, IntegralType>>
{
    using fixed_type = grotto::fixedpoint<FractionalBits, IntegralType>;

    constexpr void operator()(fixed_type & x) const
    {
        if constexpr (uses_signed_msb_v<fixed_type>)
        {
            using unsigned_type = make_unsigned_t<IntegralType>;
            auto raw = static_cast<unsigned_type>(x.integral_representation());
            raw ^= static_cast<unsigned_type>(msb_of_v<IntegralType>);
            x = grotto::make_fixed_from_integral_type<FractionalBits, IntegralType>(
                static_cast<IntegralType>(raw));
        }
    }
};

}  // namespace dpf::utils

}  // namespace dpf

namespace dpf::leaf_arithmetic
{

template <unsigned FractionalBits, typename IntegralType>
struct add_t<grotto::fixedpoint<FractionalBits, IntegralType>, simde__m128i>
{
    auto operator()(const simde__m128i & a, const simde__m128i & b) const
    {
        return add_t<IntegralType, simde__m128i>{}(a, b);
    }
};

template <unsigned FractionalBits, typename IntegralType>
struct add_t<grotto::fixedpoint<FractionalBits, IntegralType>, simde__m256i>
{
    auto operator()(const simde__m256i & a, const simde__m256i & b) const
    {
        return add_t<IntegralType, simde__m256i>{}(a, b);
    }
};

template <unsigned FractionalBits, typename IntegralType>
struct subtract_t<grotto::fixedpoint<FractionalBits, IntegralType>, simde__m128i>
{
    auto operator()(const simde__m128i & a, const simde__m128i & b) const
    {
        return subtract_t<IntegralType, simde__m128i>{}(a, b);
    }
};

template <unsigned FractionalBits, typename IntegralType>
struct subtract_t<grotto::fixedpoint<FractionalBits, IntegralType>, simde__m256i>
{
    auto operator()(const simde__m256i & a, const simde__m256i & b) const
    {
        return subtract_t<IntegralType, simde__m256i>{}(a, b);
    }
};

template <unsigned FractionalBits, typename IntegralType>
struct multiply_t<grotto::fixedpoint<FractionalBits, IntegralType>, simde__m128i>
{
    auto operator()(const simde__m128i & a,
        grotto::fixedpoint<FractionalBits, IntegralType> b) const
    {
        return multiply_t<IntegralType, simde__m128i>{}(a, b.integral_representation());
    }
};

template <unsigned FractionalBits, typename IntegralType>
struct multiply_t<grotto::fixedpoint<FractionalBits, IntegralType>, simde__m256i>
{
    auto operator()(const simde__m256i & a,
        grotto::fixedpoint<FractionalBits, IntegralType> b) const
    {
        return multiply_t<IntegralType, simde__m256i>{}(a, b.integral_representation());
    }
};

}  // namespace dpf::leaf_arithmetic

namespace std
{

template <unsigned FractionalBits,
          typename IntegralType>
class numeric_limits<grotto::fixedpoint<FractionalBits, IntegralType>>
{
    using T = grotto::fixedpoint<FractionalBits, IntegralType>;
    using I = IntegralType;
    static constexpr bool signed_rep =
        std::is_signed_v<I> || std::is_same_v<I, simde_int128>;
    static constexpr int bitwidth =
        static_cast<int>(dpf::utils::bitlength_of_v<I>);

    static constexpr I raw_lowest() noexcept
    {
        if constexpr (signed_rep)
        {
            using U = dpf::utils::make_unsigned_t<I>;
            return static_cast<I>(U{1} << static_cast<unsigned>(bitwidth - 1));
        }
        return I{};
    }

    static constexpr I raw_max() noexcept
    {
        if constexpr (signed_rep)
        {
            using U = dpf::utils::make_unsigned_t<I>;
            return static_cast<I>(static_cast<U>(raw_lowest()) - U{1});
        }
        return static_cast<I>(~I{});
    }

  public:
    static constexpr bool is_specialized = true;
    static constexpr bool is_signed = signed_rep;
    static constexpr bool is_integer = (FractionalBits == 0);
    static constexpr bool is_exact = true;
    static constexpr bool has_infinity = false;
    static constexpr bool has_quiet_NaN = false;
    static constexpr bool has_signaling_NaN = false;
    static constexpr float_denorm_style has_denorm = denorm_absent;
    static constexpr bool has_denorm_loss = false;
    static constexpr float_round_style round_style =
        FractionalBits ? round_to_nearest : round_toward_zero;
    static constexpr bool is_iec559 = false;
    static constexpr bool is_bounded = true;
    static constexpr bool is_modulo = !signed_rep;
    static constexpr int digits = bitwidth - (signed_rep ? 1 : 0);
    static constexpr int digits10 = digits * 301 / 1000;
    static constexpr int max_digits10 = 0;
    static constexpr int radix = 2;
    static constexpr int min_exponent = 1 - static_cast<int>(FractionalBits);
    static constexpr int max_exponent =
        digits - static_cast<int>(FractionalBits) + 1;
    static constexpr int min_exponent10 = min_exponent * 301 / 1000;
    static constexpr int max_exponent10 = max_exponent * 301 / 1000;
    static constexpr bool traps = false;
    static constexpr bool tinyness_before = false;

    static constexpr T lowest() noexcept { return T::from_raw(raw_lowest()); }
    static constexpr T max() noexcept { return T::from_raw(raw_max()); }
    static constexpr T min() noexcept
    {
        if constexpr (FractionalBits == 0)
        {
            return lowest();
        }
        return T::from_raw(I{1});
    }
    static constexpr T epsilon() noexcept
    {
        return FractionalBits ? T::from_raw(I{1}) : T::from_raw(I{});
    }
    static constexpr T round_error() noexcept
    {
        return FractionalBits ? T(0.5) : T(0);
    }
    static constexpr T infinity() noexcept { return max(); }
    static constexpr T quiet_NaN() noexcept { return T::from_raw(I{}); }
    static constexpr T signalling_NaN() noexcept { return T::from_raw(I{}); }
    static constexpr T denorm_min() noexcept { return min(); }
};

template <unsigned F, typename T>
struct numeric_limits<grotto::fixedpoint<F, T> const>
  : public numeric_limits<grotto::fixedpoint<F, T>> {};

template <unsigned F, typename T>
struct numeric_limits<grotto::fixedpoint<F, T> volatile>
  : public numeric_limits<grotto::fixedpoint<F, T>> {};

template <unsigned F, typename T>
struct numeric_limits<grotto::fixedpoint<F, T> const volatile>
  : public numeric_limits<grotto::fixedpoint<F, T>> {};

}  // namespace std

#include "grotto/fixedpoint_mul.hpp"

#endif  // LIBDPF_INCLUDE_GROTTO_FIXEDPOINT_HPP__