zmij/

lib.rs

//! [![github]](https://github.com/dtolnay/zmij) [![crates-io]](https://crates.io/crates/zmij) [![docs-rs]](https://docs.rs/zmij)
//!
//! [github]: https://img.shields.io/badge/github-8da0cb?style=for-the-badge&labelColor=555555&logo=github
//! [crates-io]: https://img.shields.io/badge/crates.io-fc8d62?style=for-the-badge&labelColor=555555&logo=rust
//! [docs-rs]: https://img.shields.io/badge/docs.rs-66c2a5?style=for-the-badge&labelColor=555555&logo=docs.rs
//!
//! <br>
//!
//! A double-to-string conversion algorithm based on [Schubfach] and [yy].
//!
//! This Rust implementation is a line-by-line port of Victor Zverovich's
//! implementation in C++, <https://github.com/vitaut/zmij>.
//!
//! [Schubfach]: https://fmt.dev/papers/Schubfach4.pdf
//! [yy]: https://github.com/ibireme/c_numconv_benchmark/blob/master/vendor/yy_double/yy_double.c
//!
//! <br>
//!
//! # Example
//!
//! ```
//! fn main() {
//!     let mut buffer = zmij::Buffer::new();
//!     let printed = buffer.format(1.234);
//!     assert_eq!(printed, "1.234");
//! }
//! ```
//!
//! <br>
//!
//! ## Performance
//!
//! The [dtoa-benchmark] compares this library and other Rust floating point
//! formatting implementations across a range of precisions. The vertical axis
//! in this chart shows nanoseconds taken by a single execution of
//! `zmij::Buffer::new().format_finite(value)` so a lower result indicates a
//! faster library.
//!
//! [dtoa-benchmark]: https://github.com/dtolnay/dtoa-benchmark
//!
//! ![performance](https://raw.githubusercontent.com/dtolnay/zmij/master/dtoa-benchmark.png)

#![no_std]
#![doc(html_root_url = "https://docs.rs/zmij/1.0.23")]
#![deny(unsafe_op_in_unsafe_fn)]
#![allow(non_camel_case_types, non_snake_case)]
#![allow(
    clippy::blocks_in_conditions,
    clippy::cast_possible_truncation,
    clippy::cast_possible_wrap,
    clippy::cast_ptr_alignment,
    clippy::cast_sign_loss,
    clippy::doc_markdown,
    clippy::incompatible_msrv,
    clippy::items_after_statements,
    clippy::manual_ilog2,
    clippy::many_single_char_names,
    clippy::modulo_one,
    clippy::must_use_candidate,
    clippy::needless_doctest_main,
    clippy::needless_late_init,
    clippy::never_loop,
    clippy::redundant_else,
    clippy::similar_names,
    clippy::too_many_arguments,
    clippy::too_many_lines,
    clippy::unreadable_literal,
    clippy::used_underscore_items,
    clippy::while_immutable_condition,
    clippy::wildcard_imports
)]

#[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
mod stdarch_x86;
#[cfg(test)]
mod tests;
mod traits;

#[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
use crate::stdarch_x86::{
    __m128i, _mm_add_epi64, _mm_cmpgt_epi8, _mm_cvtsi128_si64, _mm_load_si128, _mm_movemask_epi8,
    _mm_mul_epu32, _mm_mulhi_epu16, _mm_mullo_epi16, _mm_or_si128, _mm_set_epi64x,
    _mm_setzero_si128, _mm_srli_epi64,
};
#[cfg(all(
    target_arch = "x86_64",
    target_feature = "sse2",
    target_feature = "sse4.1",
    not(miri)
))]
use crate::stdarch_x86::{
    _mm_insert_epi64, _mm_mullo_epi32, _mm_shuffle_epi8, _mm_srli_epi32, _mm_storeu_si128,
};
#[cfg(all(
    target_arch = "x86_64",
    target_feature = "sse2",
    not(target_feature = "sse4.1"),
    not(miri)
))]
use crate::stdarch_x86::{
    _mm_shuffle_epi32, _mm_slli_epi16, _mm_slli_epi32, _mm_srli_epi16, _mm_sub_epi16, _MM_SHUFFLE,
};
use crate::traits::Float as _;
#[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
use core::arch::aarch64::{
    int16x8_t, int32x2_t, int32x4_t, uint16x8_t, uint64x1_t, uint8x16_t, vaddq_u16, vcgtzq_s8,
    vcombine_s32, vcreate_u64, vdup_n_s32, vdupq_n_s8, vdupq_n_u8, vget_lane_u64, vget_low_u8,
    vld1q_u8, vmla_n_s32, vmlaq_n_s16, vmlaq_n_s32, vorrq_u8, vqdmulh_n_s32, vqdmulhq_n_s16,
    vqdmulhq_n_s32, vqtbl1q_u8, vreinterpret_s32_u32, vreinterpret_s32_u64, vreinterpret_u16_s32,
    vreinterpret_u32_s32, vreinterpret_u64_u8, vreinterpretq_s16_s32, vreinterpretq_s32_u32,
    vreinterpretq_s8_u8, vreinterpretq_u16_s8, vreinterpretq_u16_u8, vreinterpretq_u64_u8,
    vreinterpretq_u8_s16, vreinterpretq_u8_u64, vrev64q_u8, vsetq_lane_u64, vshll_n_u16,
    vshr_n_u32, vshrn_n_u16, vst1q_u8,
};
#[cfg(all(any(target_arch = "aarch64", target_arch = "x86_64"), not(miri)))]
use core::arch::asm;
use core::mem::{self, MaybeUninit};
use core::ops::RangeInclusive;
use core::ptr;
use core::slice;
use core::str;
#[cfg(feature = "no-panic")]
use no_panic::no_panic;

const BUFFER_SIZE: usize = 24;
const NAN: &str = "NaN";
const INFINITY: &str = "inf";
const NEG_INFINITY: &str = "-inf";

// Declares struct members that must live in memory on ARM64 but are encoded as
// immediates in the x64 assembly.
struct AArch64Mem<const VALUE: u64> {
    #[cfg(target_arch = "aarch64")]
    value: u64,
}

impl<const VALUE: u64> AArch64Mem<VALUE> {
    const fn new() -> Self {
        AArch64Mem {
            #[cfg(target_arch = "aarch64")]
            value: VALUE,
        }
    }

    #[cfg_attr(not(target_arch = "aarch64"), allow(clippy::unused_self))]
    const fn get(&self) -> u64 {
        #[cfg(target_arch = "aarch64")]
        {
            self.value
        }

        #[cfg(not(target_arch = "aarch64"))]
        {
            VALUE
        }
    }
}

#[derive(Copy, Clone)]
#[cfg_attr(test, derive(Debug, PartialEq))]
struct uint128 {
    hi: u64,
    lo: u64,
}

// Use umul128_hi64 for division.
const USE_UMUL128_HI64: bool = cfg!(target_vendor = "apple");

// Computes 128-bit result of multiplication of two 64-bit unsigned integers.
const fn umul128(x: u64, y: u64) -> u128 {
    x as u128 * y as u128
}

#[inline]
const fn umul128_hi64(x: u64, y: u64) -> u64 {
    (umul128(x, y) >> 64) as u64
}

// Returns (x * y + c) >> 64.
#[cfg_attr(feature = "no-panic", no_panic)]
fn umul128_add_hi64(x: u64, y: u64, c: u64) -> u64 {
    ((u128::from(x) * u128::from(y) + u128::from(c)) >> 64) as u64
}

#[cfg_attr(feature = "no-panic", no_panic)]
fn umul192_hi128(x_hi: u64, x_lo: u64, y: u64) -> uint128 {
    let p = umul128(x_hi, y);
    let lo = (p as u64).wrapping_add((umul128(x_lo, y) >> 64) as u64);
    uint128 {
        hi: (p >> 64) as u64 + u64::from(lo < p as u64),
        lo,
    }
}

// Returns x / 10 for x <= 2**62.
#[cfg_attr(feature = "no-panic", no_panic)]
fn div10(x: u64) -> u64 {
    debug_assert!(x < (1 << 62));
    // ceil(2**64 / 10) computed as (1 << 63) / 5 + 1 to avoid int128.
    const DIV10_SIG64: u64 = (1 << 63) / 5 + 1;
    umul128_hi64(x, DIV10_SIG64)
}

// Computes the decimal exponent as floor(log10(2**bin_exp)) if regular or
// floor(log10(3/4 * 2**bin_exp)) otherwise, without branching.
const fn compute_dec_exp(bin_exp: i32, regular: bool) -> i32 {
    debug_assert!(bin_exp >= -1334 && bin_exp <= 2620);
    // log10_3_over_4_sig = -log10(3/4) * 2**log10_2_exp rounded to a power of 2
    const LOG10_3_OVER_4_SIG: i32 = 131_072;
    // log10_2_sig = round(log10(2) * 2**log10_2_exp)
    const LOG10_2_SIG: i32 = 315_653;
    const LOG10_2_EXP: i32 = 20;
    (bin_exp * LOG10_2_SIG - !regular as i32 * LOG10_3_OVER_4_SIG) >> LOG10_2_EXP
}

trait FloatTraits: traits::Float {
    // Note: Rust port uses wider fixed-notation ranges than upstream.
    const FIXED_DEC_EXP: RangeInclusive<i32>;

    const NUM_BITS: i32;
    const NUM_SIG_BITS: i32 = Self::MANTISSA_DIGITS as i32 - 1;
    const NUM_EXP_BITS: i32 = Self::NUM_BITS - Self::NUM_SIG_BITS - 1;
    const EXP_MASK: i32 = (1 << Self::NUM_EXP_BITS) - 1;
    const EXP_BIAS: i32 = (1 << (Self::NUM_EXP_BITS - 1)) - 1;
    const EXP_OFFSET: i32 = Self::EXP_BIAS + Self::NUM_SIG_BITS;

    type SigType: traits::UInt;
    const IMPLICIT_BIT: Self::SigType;

    type DecDigitsType: Copy;

    #[cfg(any(
        all(target_arch = "aarch64", target_feature = "neon", not(miri)),
        all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)),
    ))]
    type DecUnshuffledType;

    fn to_bits(self) -> Self::SigType;

    fn is_negative(bits: Self::SigType) -> bool {
        (bits >> (Self::NUM_BITS - 1)) != Self::SigType::from(0)
    }

    fn get_sig(bits: Self::SigType) -> Self::SigType {
        bits & (Self::IMPLICIT_BIT - Self::SigType::from(1))
    }

    fn get_exp(bits: Self::SigType) -> i64 {
        (bits << 1u8 >> (Self::NUM_SIG_BITS + 1)).into() as i64
    }

    // Converts a significand to a string, removing trailing zeros. value has up
    // to 17 decimal digits (16-17 for normals) for f64 and up to 9 digits (8-9
    // for normals) for f32.
    fn to_digits(value: u64, d: &Data) -> DecDigits<Self>;

    unsafe fn write_exp_float_simd(
        buffer: *mut u8,
        dig: &DecDigits<Self>,
        last_digit: i32,
        has_last_digit: bool,
        has_extra_digit: bool,
        exp_data: u64,
        d: &Data,
    ) -> *mut u8;
}

impl FloatTraits for f32 {
    // Upstream uses -4..=6.
    const FIXED_DEC_EXP: RangeInclusive<i32> = -6..=12;

    const NUM_BITS: i32 = 32;
    const IMPLICIT_BIT: u32 = 1 << Self::NUM_SIG_BITS;

    type SigType = u32;

    type DecDigitsType = u64;

    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    type DecUnshuffledType = uint8x16_t;
    #[cfg(all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)))]
    type DecUnshuffledType = __m128i;

    #[inline]
    fn to_bits(self) -> Self::SigType {
        self.to_bits()
    }

    #[inline]
    fn to_digits(value: u64, d: &Data) -> DecDigits<Self> {
        to_digits_32(value, d)
    }

    #[inline]
    unsafe fn write_exp_float_simd(
        buffer: *mut u8,
        dig: &DecDigits<Self>,
        last_digit: i32,
        has_last_digit: bool,
        has_extra_digit: bool,
        exp_data: u64,
        d: &Data,
    ) -> *mut u8 {
        unsafe {
            write_exp_float_simd_32(
                buffer,
                dig,
                last_digit,
                has_last_digit,
                has_extra_digit,
                exp_data,
                d,
            )
        }
    }
}

impl FloatTraits for f64 {
    // Upstream uses -4..=15.
    const FIXED_DEC_EXP: RangeInclusive<i32> = -5..=15;

    const NUM_BITS: i32 = 64;
    const IMPLICIT_BIT: u64 = 1 << Self::NUM_SIG_BITS;

    type SigType = u64;

    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    type DecDigitsType = uint16x8_t;
    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    type DecDigitsType = __m128i;
    #[cfg(not(any(
        all(target_arch = "aarch64", target_feature = "neon", not(miri)),
        all(target_arch = "x86_64", target_feature = "sse2", not(miri)),
    )))]
    type DecDigitsType = [u64; 2];

    #[cfg(any(
        all(target_arch = "aarch64", target_feature = "neon", not(miri)),
        all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)),
    ))]
    type DecUnshuffledType = ();

    #[inline]
    fn to_bits(self) -> Self::SigType {
        self.to_bits()
    }

    #[inline]
    fn to_digits(value: u64, d: &Data) -> DecDigits<Self> {
        to_digits_64(value, d)
    }

    #[inline]
    unsafe fn write_exp_float_simd(
        _buffer: *mut u8,
        _dig: &DecDigits<Self>,
        _last_digit: i32,
        _has_last_digit: bool,
        _has_extra_digit: bool,
        _exp_data: u64,
        _d: &Data,
    ) -> *mut u8 {
        ptr::null_mut()
    }
}

#[rustfmt::skip]
const POW10_MINOR: [u64; 28] = [
    0x8000000000000000, 0xa000000000000000, 0xc800000000000000,
    0xfa00000000000000, 0x9c40000000000000, 0xc350000000000000,
    0xf424000000000000, 0x9896800000000000, 0xbebc200000000000,
    0xee6b280000000000, 0x9502f90000000000, 0xba43b74000000000,
    0xe8d4a51000000000, 0x9184e72a00000000, 0xb5e620f480000000,
    0xe35fa931a0000000, 0x8e1bc9bf04000000, 0xb1a2bc2ec5000000,
    0xde0b6b3a76400000, 0x8ac7230489e80000, 0xad78ebc5ac620000,
    0xd8d726b7177a8000, 0x878678326eac9000, 0xa968163f0a57b400,
    0xd3c21bcecceda100, 0x84595161401484a0, 0xa56fa5b99019a5c8,
    0xcecb8f27f4200f3a,
];

#[rustfmt::skip]
const POW10_MAJOR: [uint128; 23] = [
    uint128 { hi: 0xaf8e5410288e1b6f, lo: 0x07ecf0ae5ee44dda }, // -303
    uint128 { hi: 0xb1442798f49ffb4a, lo: 0x99cd11cfdf41779d }, // -275
    uint128 { hi: 0xb2fe3f0b8599ef07, lo: 0x861fa7e6dcb4aa15 }, // -247
    uint128 { hi: 0xb4bca50b065abe63, lo: 0x0fed077a756b53aa }, // -219
    uint128 { hi: 0xb67f6455292cbf08, lo: 0x1a3bc84c17b1d543 }, // -191
    uint128 { hi: 0xb84687c269ef3bfb, lo: 0x3d5d514f40eea742 }, // -163
    uint128 { hi: 0xba121a4650e4ddeb, lo: 0x92f34d62616ce413 }, // -135
    uint128 { hi: 0xbbe226efb628afea, lo: 0x890489f70a55368c }, // -107
    uint128 { hi: 0xbdb6b8e905cb600f, lo: 0x5400e987bbc1c921 }, //  -79
    uint128 { hi: 0xbf8fdb78849a5f96, lo: 0xde98520472bdd034 }, //  -51
    uint128 { hi: 0xc16d9a0095928a27, lo: 0x75b7053c0f178294 }, //  -23
    uint128 { hi: 0xc350000000000000, lo: 0x0000000000000000 }, //    5
    uint128 { hi: 0xc5371912364ce305, lo: 0x6c28000000000000 }, //   33
    uint128 { hi: 0xc722f0ef9d80aad6, lo: 0x424d3ad2b7b97ef6 }, //   61
    uint128 { hi: 0xc913936dd571c84c, lo: 0x03bc3a19cd1e38ea }, //   89
    uint128 { hi: 0xcb090c8001ab551c, lo: 0x5cadf5bfd3072cc6 }, //  117
    uint128 { hi: 0xcd036837130890a1, lo: 0x36dba887c37a8c10 }, //  145
    uint128 { hi: 0xcf02b2c21207ef2e, lo: 0x94f967e45e03f4bc }, //  173
    uint128 { hi: 0xd106f86e69d785c7, lo: 0xe13336d701beba52 }, //  201
    uint128 { hi: 0xd31045a8341ca07c, lo: 0x1ede48111209a051 }, //  229
    uint128 { hi: 0xd51ea6fa85785631, lo: 0x552a74227f3ea566 }, //  257
    uint128 { hi: 0xd732290fbacaf133, lo: 0xa97c177947ad4096 }, //  285
    uint128 { hi: 0xd94ad8b1c7380874, lo: 0x18375281ae7822bc }, //  313
];

#[rustfmt::skip]
const POW10_FIXUPS: [u32; 20] = [
    0x0a4e363f, 0x00001840, 0x00006400, 0x24200040, 0x00000000,
    0x0c000000, 0x82c81380, 0x5e4ce01f, 0xd730f60f, 0x0000001b,
    0x00000000, 0xcdf7fffc, 0x6e8201d8, 0x40cd3fd1, 0xdb642501,
    0x00000d0d, 0x14042400, 0x53713840, 0x11781db4, 0x00000000,
];

// 128-bit significands of powers of 10 rounded down.
#[repr(C, align(64))]
struct Pow10SignificandTable {
    data: [u64; if Self::COMPRESS {
        0
    } else {
        Self::NUM_POW10S * 2
    }],
}

impl Pow10SignificandTable {
    const COMPRESS: bool = cfg!(opt_level = "s");
    const SPLIT_TABLES: bool = !Self::COMPRESS && cfg!(target_arch = "aarch64");
    const NUM_POW10S: usize = 618;

    // Computes the 128-bit significand of 10**i using method by Dougall Johnson.
    #[inline]
    const fn compute(i: u32) -> uint128 {
        const STRIDE: u32 = POW10_MINOR.len() as u32;
        let m = unsafe { *POW10_MINOR.as_ptr().add(((i + 10) % STRIDE) as usize) };
        let h = unsafe { *POW10_MAJOR.as_ptr().add(((i + 10) / STRIDE) as usize) };

        let h1 = umul128_hi64(h.lo, m);

        let c0 = h.lo.wrapping_mul(m);
        let c1 = h1.wrapping_add(h.hi.wrapping_mul(m));
        let c2 = (c1 < h1) as u64 + umul128_hi64(h.hi, m);

        let mut result = if (c2 >> 63) != 0 {
            uint128 { hi: c2, lo: c1 }
        } else {
            uint128 {
                hi: (c2 << 1) | (c1 >> 63),
                lo: (c1 << 1) | (c0 >> 63),
            }
        };
        result.lo -=
            ((unsafe { *POW10_FIXUPS.as_ptr().add((i >> 5) as usize) } >> (i & 31)) & 1) as u64;
        result
    }

    const fn new() -> Self {
        let mut data = [0; if Self::COMPRESS {
            0
        } else {
            Self::NUM_POW10S * 2
        }];

        let mut i = 0;
        while i < Self::NUM_POW10S && !Self::COMPRESS {
            let result = Self::compute(i as u32);
            if Self::SPLIT_TABLES {
                data[Self::NUM_POW10S - i - 1] = result.hi;
                data[Self::NUM_POW10S * 2 - i - 1] = result.lo;
            } else {
                data[i * 2] = result.hi;
                data[i * 2 + 1] = result.lo;
            }
            i += 1;
        }

        Pow10SignificandTable { data }
    }

    #[inline]
    unsafe fn get_unchecked(&self, dec_exp: i32) -> uint128 {
        const DEC_EXP_MIN: i32 = -293;
        let i = dec_exp - DEC_EXP_MIN;
        if Self::COMPRESS {
            return Self::compute(i as u32);
        }
        if !Self::SPLIT_TABLES {
            let p = unsafe { self.data.as_ptr().add((i * 2) as usize) };
            return uint128 {
                hi: unsafe { *p },
                lo: unsafe { *p.add(1) },
            };
        }

        unsafe {
            // The caller passes -e - 1 as dec_exp, so ~dec_exp recovers e.
            // Picking the base so that e itself is the index lets both loads
            // share sxtw addressing.
            #[cfg_attr(
                not(all(any(target_arch = "x86_64", target_arch = "aarch64"), not(miri))),
                allow(unused_mut)
            )]
            let mut p = self
                .data
                .as_ptr()
                .offset(Self::NUM_POW10S as isize + DEC_EXP_MIN as isize);
            #[cfg(all(any(target_arch = "x86_64", target_arch = "aarch64"), not(miri)))]
            asm!("/*{0}*/", inout(reg) p);
            uint128 {
                hi: *p.offset(!(dec_exp as isize)),
                lo: *p.offset(!(dec_exp as isize) + Self::NUM_POW10S as isize),
            }
        }
    }

    #[cfg(test)]
    fn get(&self, dec_exp: i32) -> uint128 {
        const DEC_EXP_MIN: i32 = -292;
        assert!((DEC_EXP_MIN..DEC_EXP_MIN + Self::NUM_POW10S as i32).contains(&dec_exp));
        unsafe { self.get_unchecked(dec_exp) }
    }
}

// Computes a shift so that, after scaling by a power of 10, the intermediate
// result always has a fixed 128-bit fractional part (for double).
//
// Different binary exponents can map to the same decimal exponent, but place
// the decimal point at different bit positions. The shift compensates for this.
//
// For example, both 3 * 2**59 and 3 * 2**60 have dec_exp = 2, but dividing by
// 10^dec_exp puts the decimal point in different bit positions:
//   3 * 2**59 / 100 = 1.72...e+16  (needs shift = 1 + 1)
//   3 * 2**60 / 100 = 3.45...e+16  (needs shift = 2 + 1)
#[inline]
const fn compute_exp_shift(bin_exp: i32, dec_exp: i32) -> u8 {
    debug_assert!(dec_exp >= -350 && dec_exp <= 350);
    // log2_pow10_sig = round(log2(10) * 2**log2_pow10_exp) + 1
    const LOG2_POW10_SIG: i32 = 217_707;
    const LOG2_POW10_EXP: i32 = 16;
    // pow10_bin_exp = floor(log2(10**-dec_exp))
    let pow10_bin_exp = (-dec_exp * LOG2_POW10_SIG) >> LOG2_POW10_EXP;
    // pow10 = ((pow10_hi << 64) | pow10_lo) * 2**(pow10_bin_exp - 127)
    (bin_exp + pow10_bin_exp + 1) as u8
}

struct ExpShiftTable {
    data: [u8; if Self::ENABLE {
        f64::EXP_MASK as usize + 1
    } else {
        0
    }],
}

impl ExpShiftTable {
    const ENABLE: bool = cfg!(not(opt_level = "s"));
    // extra_shift must be >= 3 to keep shift non-negative and <= 11 to fit the
    // significand into 64 bits after the shift.
    const EXTRA_SHIFT: usize = 6;

    const fn new() -> Self {
        let mut data = [0u8; if Self::ENABLE {
            f64::EXP_MASK as usize + 1
        } else {
            0
        }];

        let mut raw_exp = 0;
        while raw_exp < data.len() && Self::ENABLE {
            let mut bin_exp = raw_exp as i32 - f64::EXP_OFFSET;
            if raw_exp == 0 {
                bin_exp += 1;
            }
            let dec_exp = compute_dec_exp(bin_exp, true);
            data[raw_exp] =
                compute_exp_shift(bin_exp, dec_exp + 1).wrapping_add(Self::EXTRA_SHIFT as u8);
            raw_exp += 1;
        }

        ExpShiftTable { data }
    }
}

// An optional table of precomputed exponent strings for exponential notation.
// Each entry packs "e+dd" or "e+ddd" into a u64 with the length in byte 7.
struct ExpStringTable {
    data: [u64; if Self::ENABLE {
        (f64::MAX_10_EXP - Self::MIN_DEC_EXP + 1) as usize
    } else {
        0
    }],
}

impl ExpStringTable {
    const ENABLE: bool = cfg!(not(opt_level = "s"));
    const MIN_DEC_EXP: i32 = f64::MIN_10_EXP - f64::MAX_DIGITS10 as i32;
    const OFFSET: i32 = -Self::MIN_DEC_EXP;

    const fn new() -> Self {
        let mut data = [0u64; if Self::ENABLE {
            (f64::MAX_10_EXP - Self::MIN_DEC_EXP + 1) as usize
        } else {
            0
        }];

        let mut e = Self::MIN_DEC_EXP;
        while e <= f64::MAX_10_EXP && Self::ENABLE {
            let abs_e = e.unsigned_abs() as u64;
            let mut val = abs_e % 10 + b'0' as u64;
            if abs_e >= 10 {
                val = (val << 8) | (abs_e / 10 % 10 + b'0' as u64);
            }
            if abs_e >= 100 {
                val = (val << 8) | (abs_e / 100 + b'0' as u64);
            }
            let len = 3 + (abs_e >= 10) as u64 + (abs_e >= 100) as u64;
            data[(e + Self::OFFSET) as usize] = (len << 48)
                | (val << 16)
                | (if e >= 0 { b'+' as u64 } else { b'-' as u64 } << 8)
                | b'e' as u64;
            e += 1;
        }

        ExpStringTable { data }
    }
}

// Shuffle vectors to build strings for exponential notation.
//
// Byte positions in the source register assembled by write_exp_float_simd:
//   bytes [0, exp_pos):              BCD ASCII digits (reversed)
//   bytes [exp_pos, exp_pos + 4):    exponent string "e±NN"
//   byte  last_digit_pos:            rounded last digit
//   byte  point_pos:                 '.'
//
// The shuffle length (max 14) is stored in byte 15; the corresponding output
// byte is past the string and ignored by the caller.
#[repr(C, align(16))]
struct ExpFloatShuffleTable {
    data: [u8; if Self::ENABLE { 32 * 16 } else { 0 }],
}

struct ExpFloatShuffleTableEntry {
    #[cfg_attr(
        not(any(
            all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)),
            all(target_arch = "aarch64", target_feature = "neon", not(miri)),
        )),
        allow(dead_code)
    )]
    shuffle: *const u8,
    length: u8,
}

impl ExpFloatShuffleTable {
    const ENABLE: bool = cfg!(any(
        all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)),
        all(target_arch = "aarch64", target_feature = "neon", not(miri)),
    )) && ExpStringTable::ENABLE;

    const EXP_POS: u8 = 8;
    const LAST_DIGIT_POS: u8 = 12;
    const POINT_POS: u8 = 13;

    unsafe fn get_entry(
        &self,
        num_digits: i32,
        has_last_digit: bool,
        has_extra_digit: bool,
    ) -> ExpFloatShuffleTableEntry {
        let idx = (num_digits - 1) * 4 + i32::from(has_last_digit) * 2 + i32::from(has_extra_digit);
        ExpFloatShuffleTableEntry {
            shuffle: unsafe { self.data.as_ptr().add(idx as usize * 16) },
            length: *unsafe { self.data.get_unchecked(idx as usize * 16 + 15) },
        }
    }

    const fn new() -> Self {
        let mut data = [0u8; if Self::ENABLE { 32 * 16 } else { 0 }];

        let mut idx = 0;
        while idx < 32 && Self::ENABLE {
            let num_digits = (idx >> 2) + 1;
            let has_last_digit = ((idx >> 1) & 1) != 0;
            let has_extra_digit = (idx & 1) != 0;

            let out = idx * 16;
            let mut i = 0;
            while i < 16 {
                data[out + i] = 0x80; // shuffle high bit: output 0
                i += 1;
            }
            let leading_digit_pos = if has_extra_digit { 7 } else { 6 };
            let mut length = 0;
            if has_last_digit {
                // Always 8 BCD chars in the significand plus a last-digit char;
                // for !has_extra_digit the leading '0' of the 8-digit padded
                // BCD is shown.
                data[out + length] = leading_digit_pos;
                length += 1;
                data[out + length] = Self::POINT_POS;
                length += 1;
                let mut i = leading_digit_pos - 1;
                loop {
                    data[out + length] = i;
                    length += 1;
                    if i == 0 {
                        break;
                    }
                    i -= 1;
                }
                data[out + length] = Self::LAST_DIGIT_POS;
                length += 1;
            } else {
                length = num_digits + has_extra_digit as usize;
                // Drop the '.' for single-digit output: "5e+02", not "5.0e+02".
                if length == 2 {
                    length = 1;
                }
                data[out] = leading_digit_pos;
                data[out + 1] = Self::POINT_POS;
                let mut i = 2;
                while i < length {
                    data[out + i] = leading_digit_pos + 1 - i as u8;
                    i += 1;
                }
            }
            let mut i = 0;
            while i < 4 {
                data[out + length] = Self::EXP_POS + i;
                length += 1;
                i += 1;
            }
            data[out + 15] = length as u8;
            idx += 1;
        }

        ExpFloatShuffleTable { data }
    }
}

#[cfg(any(
    not(any(
        all(target_arch = "aarch64", target_feature = "neon", not(miri)),
        all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)),
    )),
    all(test, target_endian = "little"),
))]
#[cfg_attr(feature = "no-panic", no_panic)]
fn count_trailing_nonzeros(x: u64) -> usize {
    // We count the number of bytes until there are only zeros left.
    // The code is equivalent to
    //    8 - x.leading_zeros() / 8
    // but if the BSR instruction is emitted (as gcc on x64 does with default
    // settings), subtracting the constant before dividing allows the compiler
    // to combine it with the subtraction which it inserts due to BSR counting
    // in the opposite direction.
    //
    // Additionally, the BSR instruction requires a zero check. Since the high
    // bit is unused we can avoid the zero check by shifting the datum left by
    // one and inserting a sentinel bit at the end. This can be faster than the
    // automatically inserted range check.
    (70 - ((x.to_le() << 1) | 1).leading_zeros() as usize) / 8
}

// Align data since unaligned access may be slower when crossing a
// hardware-specific boundary.
#[repr(C, align(2))]
struct Digits2([u8; 200]);

static DIGITS2: Digits2 = Digits2(
    *b"0001020304050607080910111213141516171819\
       2021222324252627282930313233343536373839\
       4041424344454647484950515253545556575859\
       6061626364656667686970717273747576777879\
       8081828384858687888990919293949596979899",
);

// Converts value in the range [0, 100) to a string. GCC generates a bit better
// code when value is pointer-size (https://www.godbolt.org/z/5fEPMT1cc).
#[cfg_attr(feature = "no-panic", no_panic)]
unsafe fn digits2(value: usize) -> &'static u16 {
    debug_assert!(value < 100);

    #[allow(clippy::cast_ptr_alignment)]
    unsafe {
        &*DIGITS2.0.as_ptr().cast::<u16>().add(value)
    }
}

const DIV10K_EXP: i32 = 40;
const DIV10K_SIG: u32 = ((1u64 << DIV10K_EXP) / 10000 + 1) as u32;
const NEG10K: u32 = ((1u64 << 32) - 10000) as u32;

const DIV100_EXP: i32 = 19;
const DIV100_SIG: u32 = (1 << DIV100_EXP) / 100 + 1;
#[cfg(not(all(
    target_arch = "x86_64",
    target_feature = "sse2",
    not(target_feature = "sse4.1"),
    not(miri)
)))]
const NEG100: u32 = (1 << 16) - 100;

#[cfg(not(any(
    all(target_arch = "x86_64", target_feature = "sse2", not(miri)),
    all(target_arch = "aarch64", target_feature = "neon", not(miri)),
)))]
const DIV10_EXP: i32 = 10;
#[cfg(not(any(
    all(target_arch = "x86_64", target_feature = "sse2", not(miri)),
    all(target_arch = "aarch64", target_feature = "neon", not(miri)),
)))]
const DIV10_SIG: u32 = (1 << DIV10_EXP) / 10 + 1;
#[cfg(not(all(target_arch = "x86_64", target_feature = "sse2", not(miri))))]
const NEG10: u32 = (1 << 8) - 10;

const ZEROS: u64 = 0x0101010101010101 * b'0' as u64;

#[repr(C, align(64))]
struct Data {
    threshold: AArch64Mem<1_000_000_000_000_000>,
    // +6 is needed for boundary cases found by verify.py.
    biased_half: AArch64Mem<{ (1 << 63) + 6 }>,

    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    mul_const: u64,
    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    hundred_million: u64,
    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    multipliers32: int32x4_t,
    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    multipliers16: int16x8_t,

    // Ordered so that the values used to format floats fit in a single cache
    // line.
    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    div100: u128,
    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    div10: u128,
    #[cfg(all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)))]
    neg100: u128,
    #[cfg(all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)))]
    neg10: u128,
    #[cfg(all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)))]
    bswap: u128,
    #[cfg(all(
        target_arch = "x86_64",
        target_feature = "sse2",
        not(target_feature = "sse4.1"),
        not(miri)
    ))]
    hundred: u128,
    #[cfg(all(
        target_arch = "x86_64",
        target_feature = "sse2",
        not(target_feature = "sse4.1"),
        not(miri)
    ))]
    moddiv10: u128,
    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    div10k: u128,
    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    neg10k: u128,
    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    zeros: u128,

    exp_shifts: ExpShiftTable,
    exp_strings: ExpStringTable,
    pow10_significands: Pow10SignificandTable,
    exp_float_shuffles: ExpFloatShuffleTable,
}

impl Data {
    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    const fn splat64(x: u64) -> u128 {
        ((x as u128) << 64) | x as u128
    }

    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    const fn splat32(x: u32) -> u128 {
        Self::splat64(((x as u64) << 32) | x as u64)
    }

    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    const fn splat16(x: u16) -> u128 {
        Self::splat32(((x as u32) << 16) | x as u32)
    }

    #[cfg(all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)))]
    const fn pack8(a: u8, b: u8, c: u8, d: u8, e: u8, f: u8, g: u8, h: u8) -> u64 {
        ((h as u64) << 56)
            | ((g as u64) << 48)
            | ((f as u64) << 40)
            | ((e as u64) << 32)
            | ((d as u64) << 24)
            | ((c as u64) << 16)
            | ((b as u64) << 8)
            | a as u64
    }

    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    const NEG10K: i32 = 0x10000 - 10000;
}

static STATIC_DATA: Data = Data {
    threshold: AArch64Mem::new(),
    biased_half: AArch64Mem::new(),

    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    mul_const: 0xabcc77118461cefd,
    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    hundred_million: 100000000,
    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    multipliers32: unsafe {
        mem::transmute::<[i32; 4], int32x4_t>([
            DIV10K_SIG as i32,
            Data::NEG10K,
            (DIV100_SIG << 12) as i32,
            NEG100 as i32,
        ])
    },
    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    multipliers16: unsafe {
        mem::transmute::<[i16; 8], int16x8_t>([0xce0, NEG10 as i16, 0, 0, 0, 0, 0, 0])
    },

    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    div100: Data::splat32(DIV100_SIG),
    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    div10: Data::splat16(((1u32 << 16) / 10 + 1) as u16),
    #[cfg(all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)))]
    neg100: Data::splat32(NEG100),
    #[cfg(all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)))]
    neg10: Data::splat16((1 << 8) - 10),
    #[cfg(all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)))]
    bswap: Data::pack8(15, 14, 13, 12, 11, 10, 9, 8) as u128
        | (Data::pack8(7, 6, 5, 4, 3, 2, 1, 0) as u128) << 64,
    #[cfg(all(
        target_arch = "x86_64",
        target_feature = "sse2",
        not(target_feature = "sse4.1"),
        not(miri)
    ))]
    hundred: Data::splat32(100),
    #[cfg(all(
        target_arch = "x86_64",
        target_feature = "sse2",
        not(target_feature = "sse4.1"),
        not(miri)
    ))]
    moddiv10: Data::splat16(10 * (1 << 8) - 1),
    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    div10k: Data::splat64(DIV10K_SIG as u64),
    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    neg10k: Data::splat64(NEG10K as u64),
    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    zeros: Data::splat64(ZEROS),

    exp_shifts: ExpShiftTable::new(),
    exp_strings: ExpStringTable::new(),
    pow10_significands: Pow10SignificandTable::new(),
    exp_float_shuffles: ExpFloatShuffleTable::new(),
};

// Converts four numbers < 10000, one in each 32-bit lane, to BCD digits.
#[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
#[cfg_attr(feature = "no-panic", no_panic)]
fn to_bcd_4x4(mut efgh_abcd_mnop_ijkl: int32x4_t, d: &Data) -> uint8x16_t {
    unsafe {
        // Compiler barrier, or clang breaks the subsequent MLA into UADDW +
        // MUL.
        asm!("/*{:v}*/", inout(vreg) efgh_abcd_mnop_ijkl);

        let ef_ab_mn_ij: int32x4_t = vqdmulhq_n_s32(
            efgh_abcd_mnop_ijkl,
            mem::transmute::<int32x4_t, [i32; 4]>(d.multipliers32)[2],
        );
        let gh_ef_cd_ab_op_mn_kl_ij: int16x8_t = vreinterpretq_s16_s32(vmlaq_n_s32(
            efgh_abcd_mnop_ijkl,
            ef_ab_mn_ij,
            mem::transmute::<int32x4_t, [i32; 4]>(d.multipliers32)[3],
        ));
        let high_10s: int16x8_t = vqdmulhq_n_s16(
            gh_ef_cd_ab_op_mn_kl_ij,
            mem::transmute::<int16x8_t, [i16; 8]>(d.multipliers16)[0],
        );
        vreinterpretq_u8_s16(vmlaq_n_s16(
            gh_ef_cd_ab_op_mn_kl_ij,
            high_10s,
            mem::transmute::<int16x8_t, [i16; 8]>(d.multipliers16)[1],
        ))
    }
}

// An optimized version for NEON by Dougall Johnson.
#[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
#[cfg_attr(feature = "no-panic", no_panic)]
#[inline]
fn to_unshuffled_digits(value: u64, d: &Data) -> uint8x16_t {
    let mut hundred_million = d.hundred_million;

    // Compiler barrier, or clang narrows the load to 32-bit and unpairs it.
    unsafe {
        asm!("/*{0}*/", inout(reg) hundred_million);
    }

    // abcdefgh = value / 100000000, ijklmnop = value % 100000000.
    let abcdefgh = (umul128(value, d.mul_const) >> 90) as u64;
    let ijklmnop = value - abcdefgh * hundred_million;

    unsafe {
        let ijklmnop_abcdefgh_64: uint64x1_t =
            mem::transmute::<u64, uint64x1_t>((ijklmnop << 32) | abcdefgh);
        let abcdefgh_ijklmnop: int32x2_t = vreinterpret_s32_u64(ijklmnop_abcdefgh_64);

        let abcd_ijkl: int32x2_t = vreinterpret_s32_u32(vshr_n_u32(
            vreinterpret_u32_s32(vqdmulh_n_s32(
                abcdefgh_ijklmnop,
                mem::transmute::<int32x4_t, [i32; 4]>(d.multipliers32)[0],
            )),
            9,
        ));
        let efgh_abcd_mnop_ijkl_32: int32x2_t = vmla_n_s32(
            abcdefgh_ijklmnop,
            abcd_ijkl,
            mem::transmute::<int32x4_t, [i32; 4]>(d.multipliers32)[1],
        );

        let efgh_abcd_mnop_ijkl: int32x4_t =
            vreinterpretq_s32_u32(vshll_n_u16(vreinterpret_u16_s32(efgh_abcd_mnop_ijkl_32), 0));

        to_bcd_4x4(efgh_abcd_mnop_ijkl, d)
    }
}

// Converts four numbers < 10000, one in each 32-bit lane, to BCD digits.
// Digits in each 32-bit lane will be in order for SSE2, reversed for SSE4.1.
#[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
#[cfg_attr(feature = "no-panic", no_panic)]
fn to_bcd_4x4(y: __m128i, d: &Data) -> __m128i {
    unsafe {
        let div100 = _mm_load_si128(ptr::addr_of!(d.div100).cast::<__m128i>());
        let div10 = _mm_load_si128(ptr::addr_of!(d.div10).cast::<__m128i>());

        #[cfg(target_feature = "sse4.1")]
        {
            let neg100 = _mm_load_si128(ptr::addr_of!(d.neg100).cast::<__m128i>());
            let neg10 = _mm_load_si128(ptr::addr_of!(d.neg10).cast::<__m128i>());

            // _mm_mullo_epi32 is SSE 4.1
            let z: __m128i = _mm_add_epi64(
                y,
                _mm_mullo_epi32(neg100, _mm_srli_epi32(_mm_mulhi_epu16(y, div100), 3)),
            );
            _mm_add_epi64(z, _mm_mullo_epi16(neg10, _mm_mulhi_epu16(z, div10)))
        }

        #[cfg(not(target_feature = "sse4.1"))]
        {
            let hundred = _mm_load_si128(ptr::addr_of!(d.hundred).cast::<__m128i>());
            let moddiv10 = _mm_load_si128(ptr::addr_of!(d.moddiv10).cast::<__m128i>());

            let y_div_100: __m128i = _mm_srli_epi16(_mm_mulhi_epu16(y, div100), 3);
            let y_mod_100: __m128i = _mm_sub_epi16(y, _mm_mullo_epi16(y_div_100, hundred));
            let z: __m128i = _mm_or_si128(_mm_slli_epi32(y_mod_100, 16), y_div_100);
            _mm_sub_epi16(
                _mm_slli_epi16(z, 8),
                _mm_mullo_epi16(moddiv10, _mm_mulhi_epu16(z, div10)),
            )
        }
    }
}

#[cfg(not(any(
    all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)),
    all(target_arch = "aarch64", target_feature = "neon", not(miri)),
)))]
struct BcdResult {
    bcd: u64,
    len: usize,
}

#[cfg(not(any(
    all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)),
    all(target_arch = "aarch64", target_feature = "neon", not(miri)),
)))]
#[cfg_attr(feature = "no-panic", no_panic)]
fn to_bcd8(abcdefgh: u64) -> BcdResult {
    #[cfg(not(all(target_arch = "x86_64", target_feature = "sse2", not(miri))))]
    let bcd = {
        // An optimization from Xiang JunBo.
        // Three steps BCD. Base 10000 -> base 100 -> base 10.
        // div and mod are evaluated simultaneously as, e.g.
        //   (abcdefgh / 10000) << 32 + (abcdefgh % 10000)
        //      == abcdefgh + (2**32 - 10000) * (abcdefgh / 10000)))
        // where the division on the RHS is implemented by the usual multiply + shift
        // trick and the fractional bits are masked away.
        let abcd_efgh =
            abcdefgh + u64::from(NEG10K) * ((abcdefgh * u64::from(DIV10K_SIG)) >> DIV10K_EXP);
        let ab_cd_ef_gh = abcd_efgh
            + u64::from(NEG100)
                * (((abcd_efgh * u64::from(DIV100_SIG)) >> DIV100_EXP) & 0x7f0000007f);
        let a_b_c_d_e_f_g_h = ab_cd_ef_gh
            + u64::from(NEG10)
                * (((ab_cd_ef_gh * u64::from(DIV10_SIG)) >> DIV10_EXP) & 0xf000f000f000f);
        a_b_c_d_e_f_g_h.to_be()
    };

    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    let bcd = {
        // Load constants from memory.
        let mut d = ptr::addr_of!(STATIC_DATA);
        let d = unsafe {
            asm!("/*{0}*/", inout(reg) d);
            &*d
        };

        // Evaluate the 4-digit limbs and arrange them such that we get a
        // result which is in the correct order.
        let abcd_efgh = (abcdefgh << 32)
            - ((10000u64 << 32) - 1) * ((abcdefgh * u64::from(DIV10K_SIG)) >> DIV10K_EXP);
        let v: __m128i = to_bcd_4x4(_mm_set_epi64x(0, abcd_efgh as i64), d);
        (unsafe { _mm_cvtsi128_si64(v) }) as u64
    };

    BcdResult {
        bcd,
        len: count_trailing_nonzeros(bcd),
    }
}

struct DecDigits<Float: FloatTraits> {
    digits: Float::DecDigitsType,
    // `unshuffled` is the byte-reversed BCD vector used by write_exp_float_simd.
    #[cfg(any(
        all(target_arch = "aarch64", target_feature = "neon", not(miri)),
        all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)),
    ))]
    unshuffled: Float::DecUnshuffledType,
    num_digits: usize,
}

#[cfg_attr(feature = "no-panic", no_panic)]
#[inline]
fn to_digits_64(value: u64, #[allow(unused_variables)] d: &Data) -> DecDigits<f64> {
    #[cfg(not(any(
        all(target_arch = "aarch64", target_feature = "neon", not(miri)),
        all(target_arch = "x86_64", target_feature = "sse2", not(miri)),
    )))]
    {
        let hi = (value / 100_000_000) as u32;
        let lo = (value % 100_000_000) as u32;
        let hi_bcd = to_bcd8(hi as u64);
        if lo == 0 {
            return DecDigits {
                digits: [hi_bcd.bcd + ZEROS, ZEROS],
                num_digits: hi_bcd.len,
            };
        }
        let lo_bcd = to_bcd8(lo as u64);
        DecDigits {
            digits: [hi_bcd.bcd + ZEROS, lo_bcd.bcd + ZEROS],
            num_digits: 8 + lo_bcd.len,
        }
    }

    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    {
        unsafe {
            let unshuffled_digits = to_unshuffled_digits(value, d);
            let digits: uint8x16_t = vrev64q_u8(unshuffled_digits);
            let str: uint16x8_t = vaddq_u16(
                vreinterpretq_u16_u8(digits),
                vreinterpretq_u16_s8(vdupq_n_s8(b'0' as i8)),
            );
            let is_not_zero: uint16x8_t =
                vreinterpretq_u16_u8(vcgtzq_s8(vreinterpretq_s8_u8(digits)));
            let nonzero_mask: u64 =
                vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(is_not_zero, 4)), 0);
            DecDigits {
                digits: str,
                unshuffled: (),
                num_digits: 16 - (nonzero_mask.leading_zeros() as usize >> 2),
            }
        }
    }

    #[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(miri)))]
    {
        let hi = (value / 100_000_000) as u32;
        let lo = (value % 100_000_000) as u32;

        unsafe {
            let div10k = _mm_load_si128(ptr::addr_of!(d.div10k).cast::<__m128i>());
            let neg10k = _mm_load_si128(ptr::addr_of!(d.neg10k).cast::<__m128i>());
            let x: __m128i = _mm_set_epi64x(i64::from(hi), i64::from(lo));
            #[cfg_attr(target_feature = "sse4.1", allow(unused_mut))]
            let mut y: __m128i = _mm_add_epi64(
                x,
                _mm_mul_epu32(neg10k, _mm_srli_epi64(_mm_mul_epu32(x, div10k), DIV10K_EXP)),
            );

            // Shuffle to ensure correctly ordered result from SSE2 path.
            #[cfg(not(target_feature = "sse4.1"))]
            {
                y = _mm_shuffle_epi32(y, _MM_SHUFFLE(0, 1, 2, 3));
            }

            #[cfg_attr(not(target_feature = "sse4.1"), allow(unused_mut))]
            let mut bcd: __m128i = to_bcd_4x4(y, d);
            let zeros = _mm_load_si128(ptr::addr_of!(d.zeros).cast::<__m128i>());

            // Computed against current bcd (rather than the post-bswap bcd) so
            // the mask is derived in parallel with the shuffle on the SSE4.1
            // path.
            let mask = _mm_movemask_epi8(_mm_cmpgt_epi8(bcd, _mm_setzero_si128())) as u64;
            // Trailing zeros are in the low bits for SSE4.1, the high bits for
            // SSE2.
            let len = if cfg!(target_feature = "sse4.1") {
                16 - mask.trailing_zeros()
            } else {
                64 - mask.leading_zeros()
            };

            #[cfg(target_feature = "sse4.1")]
            {
                bcd = _mm_shuffle_epi8(
                    bcd,
                    _mm_load_si128(ptr::addr_of!(d.bswap).cast::<__m128i>()),
                ); // SSSE3
            }

            DecDigits {
                digits: _mm_or_si128(bcd, zeros),
                #[cfg(target_feature = "sse4.1")]
                unshuffled: (),
                num_digits: len as usize,
            }
        }
    }
}

#[cfg_attr(feature = "no-panic", no_panic)]
#[inline]
fn to_digits_32(value: u64, #[allow(unused_variables)] d: &Data) -> DecDigits<f32> {
    #[cfg(all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)))]
    {
        // Inline to_bcd8's SSE4.1 body so we can return the unshuffled xmm too;
        // the exponential-notation path uses it to skip the bswap-via-gpr.
        let abcd_efgh = value + u64::from(NEG10K) * ((value * u64::from(DIV10K_SIG)) >> DIV10K_EXP);
        let bcd_xmm = to_bcd_4x4(_mm_set_epi64x(0, abcd_efgh as i64), d);
        let unshuffled_bcd = unsafe { _mm_cvtsi128_si64(bcd_xmm) } as u64;
        let len = if unshuffled_bcd != 0 {
            8 - unshuffled_bcd.trailing_zeros() / 8
        } else {
            0
        };
        DecDigits {
            digits: unshuffled_bcd.swap_bytes() + ZEROS,
            unshuffled: bcd_xmm,
            num_digits: len as usize,
        }
    }

    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    {
        // Inline to_bcd8's NEON body so we can return the unshuffled vector
        // too; the exponential-notation path uses it to skip the
        // simd->gpr->bswap->simd roundtrip needed to materialize `digits`.
        let abcd_efgh = value + u64::from(NEG10K) * ((value * u64::from(DIV10K_SIG)) >> DIV10K_EXP);
        let unshuffled: uint8x16_t = unsafe {
            let input: int32x4_t =
                vcombine_s32(vreinterpret_s32_u64(vcreate_u64(abcd_efgh)), vdup_n_s32(0));
            to_bcd_4x4(input, d)
        };
        let unshuffled_bcd =
            unsafe { vget_lane_u64(vreinterpret_u64_u8(vget_low_u8(unshuffled)), 0) };
        let len = if unshuffled_bcd != 0 {
            8 - unshuffled_bcd.trailing_zeros() / 8
        } else {
            0
        };
        DecDigits {
            digits: unshuffled_bcd.swap_bytes() + ZEROS,
            unshuffled,
            num_digits: len as usize,
        }
    }

    #[cfg(not(any(
        all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)),
        all(target_arch = "aarch64", target_feature = "neon", not(miri)),
    )))]
    {
        let result = to_bcd8(value);
        DecDigits {
            digits: result.bcd + ZEROS,
            num_digits: result.len,
        }
    }
}

#[cfg_attr(feature = "no-panic", no_panic)]
unsafe fn write_exp_float_simd_32(
    buffer: *mut u8,
    dig: &DecDigits<f32>,
    last_digit: i32,
    has_last_digit: bool,
    has_extra_digit: bool,
    exp_data: u64,
    d: &Data,
) -> *mut u8 {
    // Packed for insertion into lane 1: byte 0 of `tail` lands at register byte
    // exp_pos (8), so the exp string fills exp_pos..exp_pos+3; the prefix
    // shifts place '0'+last_digit at last_digit_pos (12) and '.' at point_pos
    // (13).
    let prefix = (u32::from(b'.') << 8) + u32::from(b'0') + last_digit as u32;
    #[cfg_attr(
        not(any(
            all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)),
            all(target_arch = "aarch64", target_feature = "neon", not(miri)),
        )),
        allow(unused_variables)
    )]
    let tail = exp_data | (u64::from(prefix) << 32);
    let entry = unsafe {
        d.exp_float_shuffles
            .get_entry(dig.num_digits as i32, has_last_digit, has_extra_digit)
    };

    #[cfg(all(target_arch = "x86_64", target_feature = "sse4.1", not(miri)))]
    unsafe {
        let ascii: __m128i = _mm_or_si128(
            dig.unshuffled,
            _mm_load_si128(ptr::addr_of!(d.zeros).cast::<__m128i>()),
        );
        let src: __m128i = _mm_insert_epi64(ascii, tail as i64, 1);
        let shuffle: __m128i = _mm_load_si128(entry.shuffle.cast::<__m128i>());
        let out: __m128i = _mm_shuffle_epi8(src, shuffle);
        _mm_storeu_si128(buffer.cast::<__m128i>(), out);
    }

    #[cfg(all(target_arch = "aarch64", target_feature = "neon", not(miri)))]
    unsafe {
        let ascii: uint8x16_t = vorrq_u8(dig.unshuffled, vdupq_n_u8(b'0'));
        let src: uint8x16_t =
            vreinterpretq_u8_u64(vsetq_lane_u64(tail, vreinterpretq_u64_u8(ascii), 1));
        let shuffle: uint8x16_t = vld1q_u8(entry.shuffle);
        let out: uint8x16_t = vqtbl1q_u8(src, shuffle);
        vst1q_u8(buffer, out);
    }

    let length = entry.length as usize - usize::from((exp_data & 0xff000000) == 0);
    unsafe { buffer.add(length) }
}

struct ToDecimalResult {
    sig: i64,
    exp: i32,
    last_digit: u8,
    has_last_digit: bool,
}

// Here be 🐉s.
// Converts a binary FP number bin_sig * 2**bin_exp to the shortest decimal
// representation, where bin_exp = raw_exp - exp_offset.
#[cfg_attr(feature = "no-panic", no_panic)]
#[inline]
fn to_decimal<Float, UInt>(bin_sig: UInt, raw_exp: i64, regular: bool, d: &Data) -> ToDecimalResult
where
    Float: FloatTraits,
    UInt: traits::UInt,
{
    let bin_exp = raw_exp - i64::from(Float::EXP_OFFSET);
    let num_bits = mem::size_of::<UInt>() as i32 * 8;
    const EXTRA_SHIFT: usize = ExpShiftTable::EXTRA_SHIFT;

    if !regular {
        let dec_exp = compute_dec_exp(bin_exp as i32, false);
        let shift = compute_exp_shift(bin_exp as i32, dec_exp + 1).wrapping_add(EXTRA_SHIFT as u8);
        let pow10 = unsafe { d.pow10_significands.get_unchecked(-dec_exp - 1) };
        let p = umul192_hi128(pow10.hi, pow10.lo, (bin_sig << shift).into());

        let mut integral = p.hi >> EXTRA_SHIFT;
        let fractional = (p.hi << (64 - EXTRA_SHIFT)) | (p.lo >> EXTRA_SHIFT);

        let half_ulp = pow10.hi >> (EXTRA_SHIFT + 1 - shift as usize);
        let round_up = half_ulp > u64::MAX - fractional;
        let round_down = (half_ulp >> 1) > fractional;
        integral += u64::from(round_up);

        let mut digit = umul128_add_hi64(fractional, 10, (1 << 63) - 1) as i32;
        let lo = umul128_add_hi64(fractional.wrapping_sub(half_ulp >> 1), 10, !0) as i32;
        if digit < lo {
            digit = lo;
        }
        return ToDecimalResult {
            sig: integral as i64,
            exp: dec_exp,
            last_digit: digit as u8,
            has_last_digit: !(round_up || round_down),
        };
    }

    const LOG10_2_SIG: u64 = 78_913;
    const LOG10_2_EXP: i32 = 18;
    #[allow(unused_mut)]
    let mut dec_exp = if USE_UMUL128_HI64 {
        umul128_hi64(bin_exp as u64, LOG10_2_SIG << (64 - LOG10_2_EXP)) as i32
    } else {
        compute_dec_exp(bin_exp as i32, true)
    };
    #[cfg(not(miri))]
    #[allow(unused_unsafe)]
    unsafe {
        // Force 32-bit reg for sxtw addressing.
        #[cfg(target_arch = "x86_64")]
        asm!("/*{0:e}*/", inout(reg) dec_exp);
        #[cfg(target_arch = "aarch64")]
        asm!("/*{0:w}*/", inout(reg) dec_exp);
    }
    let mut shift = if ExpShiftTable::ENABLE {
        *unsafe {
            d.exp_shifts
                .data
                .get_unchecked((bin_exp + i64::from(f64::EXP_OFFSET)) as usize)
        }
    } else {
        compute_exp_shift(bin_exp as i32, dec_exp + 1).wrapping_add(EXTRA_SHIFT as u8)
    };
    let even = UInt::from(1) - (bin_sig & UInt::from(1));

    if num_bits == 32 {
        const EXTRA_SHIFT: usize = 34;
        shift += (EXTRA_SHIFT - ExpShiftTable::EXTRA_SHIFT) as u8;
        let pow10_hi = unsafe { d.pow10_significands.get_unchecked(-dec_exp - 1) }.hi;
        let p = umul128_hi64(pow10_hi + 1, bin_sig.into() << shift);

        let mut integral = p >> EXTRA_SHIFT;
        let fractional = p & ((1u64 << EXTRA_SHIFT) - 1);

        let half_ulp = (pow10_hi >> (65 - shift as usize)) + even.into();
        let round_up = ((fractional + half_ulp) >> EXTRA_SHIFT) != 0;
        let round_down = half_ulp > fractional;
        integral += u64::from(round_up);

        let mut digit = ((fractional * 10 + (1u64 << (EXTRA_SHIFT - 1))) >> EXTRA_SHIFT) as i32;
        if fractional == (1u64 << (EXTRA_SHIFT - 2)) {
            digit = 2; // Round 2.5 to 2.
        }
        return ToDecimalResult {
            sig: integral as i64,
            exp: dec_exp,
            last_digit: digit as u8,
            has_last_digit: !(round_up || round_down),
        };
    }

    // An optimization by Xiang JunBo:
    // Scale by 10**(-dec_exp-1) to directly produce the shorter candidate
    // (15-16 digits), deriving the extra digit from the fractional part.
    // This eliminates div10 from the critical path.
    //
    // value = 5.0507837461e-27
    // next  = 5.0507837461000010e-27
    //
    // c = integral.fractional' = 5050783746100000.3153987... (value)
    //                            5050783746100001.0328635... (next)
    //                 half_ulp =                0.3587324...
    //
    // fractional = fractional' * 2**64 = 5818079786399166407
    //
    //    5050783746100000.0       c               upper    5050783746100001.0
    //             s              l|   L             |               S
    // ──┬────┬────┼────┬────┬────┼*───┼────┬────┬───*┬────┬────┬────┼─*──┬───
    //  .8   .9   .0   .1   .2   .3   .4   .5   .6   .7   .8   .9   .0 | .1
    //           └─────────────────┼─────────────────┘                next
    //                            1ulp
    //
    // s - shorter underestimate, S - shorter overestimate
    // l - longer underestimate,  L - longer overestimate
    let pow10 = unsafe { d.pow10_significands.get_unchecked(-dec_exp - 1) };
    let p = umul192_hi128(pow10.hi, pow10.lo, (bin_sig << shift).into());

    let mut integral = p.hi >> EXTRA_SHIFT;
    let fractional = (p.hi << (64 - EXTRA_SHIFT)) | (p.lo >> EXTRA_SHIFT);

    let half_ulp = (pow10.hi >> (EXTRA_SHIFT + 1 - shift as usize)) + even.into();
    let round_up = fractional.wrapping_add(half_ulp) < fractional;
    let round_down = half_ulp > fractional;
    integral += u64::from(round_up); // Compute integral before digit.

    // Derive the extra digit from the fractional part (parallel with rounding).
    let mut digit = umul128_add_hi64(fractional, 10, d.biased_half.get()) as i32;
    if fractional == (1u64 << 62) {
        digit = 2; // Round 2.5 to 2.
    }
    ToDecimalResult {
        sig: integral as i64,
        exp: dec_exp,
        last_digit: digit as u8,
        has_last_digit: !(round_up || round_down),
    }
}

/// Writes the shortest correctly rounded decimal representation of `value` to
/// `buffer`. `buffer` should point to a buffer of size `buffer_size` or larger.
#[cfg_attr(feature = "no-panic", no_panic)]
unsafe fn write<Float>(value: Float, mut buffer: *mut u8) -> *mut u8
where
    Float: FloatTraits,
{
    let bits = value.to_bits();
    // It is beneficial to extract exponent and significand early.
    let bin_exp = Float::get_exp(bits); // binary exponent
    let bin_sig = Float::get_sig(bits); // binary significand

    unsafe {
        *buffer = b'-';
    }
    buffer = unsafe { buffer.add(usize::from(Float::is_negative(bits))) };

    #[allow(unused_mut)]
    let mut d = ptr::addr_of!(STATIC_DATA);
    let d = unsafe {
        // Load constants from memory.
        #[cfg(all(any(target_arch = "aarch64", target_arch = "x86_64"), not(miri)))]
        asm!("/*{0}*/", inout(reg) d);
        &*d
    };
    let threshold = if Float::NUM_BITS == 64 {
        d.threshold.get()
    } else {
        10_000_000
    };

    let mut dec;
    if bin_exp == 0 {
        if bin_sig == Float::SigType::from(0) {
            return unsafe {
                *buffer = b'0';
                *buffer.add(1) = b'.';
                *buffer.add(2) = b'0';
                buffer.add(3)
            };
        }
        dec = to_decimal::<Float, Float::SigType>(bin_sig, 1, true, d);
        let mut dec_sig =
            dec.sig * 10 + (-i64::from(dec.has_last_digit) & i64::from(dec.last_digit));
        let mut dec_exp = dec.exp;
        while dec_sig < threshold as i64 {
            dec_sig *= 10;
            dec_exp -= 1;
        }
        let d = div10(dec_sig as u64);
        let last_digit = dec_sig - d as i64 * 10;
        dec = ToDecimalResult {
            sig: d as i64,
            exp: dec_exp,
            last_digit: last_digit as u8,
            has_last_digit: last_digit != 0,
        };
    } else {
        dec = to_decimal::<Float, Float::SigType>(
            bin_sig | Float::IMPLICIT_BIT,
            bin_exp,
            bin_sig != Float::SigType::from(0),
            d,
        );
    }
    let mut has_last_digit = dec.has_last_digit;
    let has_extra_digit = dec.sig >= threshold as i64;
    let mut dec_exp = dec.exp + Float::MAX_DIGITS10 as i32 - 2 + i32::from(has_extra_digit);
    if Float::NUM_BITS == 32 && dec.sig < 1_000_000 {
        dec.sig = 10 * dec.sig + (-i64::from(has_last_digit) & i64::from(dec.last_digit));
        has_last_digit = false;
        dec_exp -= 1;
    }

    // Write significand.
    let dig = Float::to_digits(dec.sig as u64, d);

    if Float::NUM_BITS == 32
        && ExpFloatShuffleTable::ENABLE
        && !Float::FIXED_DEC_EXP.contains(&dec_exp)
    {
        unsafe {
            let exp_data = *d
                .exp_strings
                .data
                .get_unchecked((dec_exp + ExpStringTable::OFFSET) as usize);
            return Float::write_exp_float_simd(
                buffer,
                &dig,
                i32::from(dec.last_digit),
                has_last_digit,
                has_extra_digit,
                exp_data,
                d,
            );
        }
    }

    let bcd_size = if Float::NUM_BITS == 64 { 16 } else { 8 };
    unsafe {
        buffer
            .add(usize::from(has_extra_digit))
            .cast::<Float::DecDigitsType>()
            .write_unaligned(dig.digits);
        buffer
            .add(usize::from(has_extra_digit) + bcd_size)
            .write(b'0' + dec.last_digit);
    }
    let length = usize::from(has_extra_digit)
        + if has_last_digit {
            bcd_size + 1
        } else {
            dig.num_digits
        }
        - 1;

    if Float::FIXED_DEC_EXP.contains(&dec_exp) {
        if length as i32 - 1 <= dec_exp {
            // 1234e7 -> 12340000000.0
            return unsafe {
                ptr::copy(buffer.add(1), buffer, length);
                ptr::write_bytes(buffer.add(length), b'0', dec_exp as usize + 3 - length);
                *buffer.add(dec_exp as usize + 1) = b'.';
                buffer.add(dec_exp as usize + 3)
            };
        } else if 0 <= dec_exp {
            // 1234e-2 -> 12.34
            return unsafe {
                ptr::copy(buffer.add(1), buffer, dec_exp as usize + 1);
                *buffer.add(dec_exp as usize + 1) = b'.';
                buffer.add(length + 1)
            };
        } else {
            // 1234e-6 -> 0.001234
            return unsafe {
                ptr::copy(buffer.add(1), buffer.add((1 - dec_exp) as usize), length);
                ptr::write_bytes(buffer, b'0', (1 - dec_exp) as usize);
                *buffer.add(1) = b'.';
                buffer.add((1 - dec_exp) as usize + length)
            };
        }
    }

    unsafe {
        // 1234e30 -> 1.234e33
        *buffer = *buffer.add(1);
        *buffer.add(1) = b'.';
    }
    buffer = unsafe { buffer.add(length + usize::from(length > 1)) };

    // Write exponent.
    if ExpStringTable::ENABLE {
        let mut exp_data = unsafe {
            *d.exp_strings
                .data
                .get_unchecked((dec_exp + ExpStringTable::OFFSET) as usize)
        };
        let len = (exp_data >> 48) as usize;
        exp_data = exp_data.to_le();
        unsafe {
            ptr::copy_nonoverlapping(
                ptr::addr_of!(exp_data).cast::<u8>(),
                buffer,
                if Float::MAX_10_EXP >= 100 { 5 } else { 4 },
            );
            return buffer.add(len);
        }
    }
    let sign_ptr = buffer;
    let e_sign = if dec_exp >= 0 {
        (u16::from(b'+') << 8) | u16::from(b'e')
    } else {
        (u16::from(b'-') << 8) | u16::from(b'e')
    };
    buffer = unsafe { buffer.add(1) };
    dec_exp = if dec_exp >= 0 { dec_exp } else { -dec_exp };
    buffer = unsafe { buffer.add(usize::from(dec_exp >= 10)) };
    if Float::MAX_10_EXP >= 100 {
        // digit = dec_exp / 100
        let digit = if USE_UMUL128_HI64 {
            umul128_hi64(dec_exp as u64, 0x290000000000000) as u32
        } else {
            (dec_exp as u32 * DIV100_SIG) >> DIV100_EXP
        };
        unsafe {
            *buffer = b'0' + digit as u8;
        }
        buffer = unsafe { buffer.add(usize::from(dec_exp >= 100)) };
        dec_exp -= (digit * 100) as i32;
    }
    unsafe {
        buffer
            .cast::<u16>()
            .write_unaligned(*digits2(dec_exp as usize));
        sign_ptr.cast::<u16>().write_unaligned(e_sign.to_le());
        buffer.add(2)
    }
}

/// Safe API for formatting floating point numbers to text.
///
/// ## Example
///
/// ```
/// let mut buffer = zmij::Buffer::new();
/// let printed = buffer.format_finite(1.234);
/// assert_eq!(printed, "1.234");
/// ```
pub struct Buffer {
    bytes: [MaybeUninit<u8>; BUFFER_SIZE],
}

impl Buffer {
    /// This is a cheap operation; you don't need to worry about reusing buffers
    /// for efficiency.
    #[inline]
    #[cfg_attr(feature = "no-panic", no_panic)]
    pub fn new() -> Self {
        let bytes = [MaybeUninit::<u8>::uninit(); BUFFER_SIZE];
        Buffer { bytes }
    }

    /// Print a floating point number into this buffer and return a reference to
    /// its string representation within the buffer.
    ///
    /// # Special cases
    ///
    /// This function formats NaN as the string "NaN", positive infinity as
    /// "inf", and negative infinity as "-inf" to match std::fmt.
    ///
    /// If your input is known to be finite, you may get better performance by
    /// calling the `format_finite` method instead of `format` to avoid the
    /// checks for special cases.
    #[cfg_attr(feature = "no-panic", no_panic)]
    pub fn format<F: Float>(&mut self, f: F) -> &str {
        if f.is_nonfinite() {
            f.format_nonfinite()
        } else {
            self.format_finite(f)
        }
    }

    /// Print a floating point number into this buffer and return a reference to
    /// its string representation within the buffer.
    ///
    /// # Special cases
    ///
    /// This function **does not** check for NaN or infinity. If the input
    /// number is not a finite float, the printed representation will be some
    /// correctly formatted but unspecified numerical value.
    ///
    /// Please check [`is_finite`] yourself before calling this function, or
    /// check [`is_nan`] and [`is_infinite`] and handle those cases yourself.
    ///
    /// [`is_finite`]: f64::is_finite
    /// [`is_nan`]: f64::is_nan
    /// [`is_infinite`]: f64::is_infinite
    #[cfg_attr(feature = "no-panic", no_panic)]
    pub fn format_finite<F: Float>(&mut self, f: F) -> &str {
        unsafe {
            let end = f.write_to_zmij_buffer(self.bytes.as_mut_ptr().cast::<u8>());
            let len = end.offset_from(self.bytes.as_ptr().cast::<u8>()) as usize;
            let slice = slice::from_raw_parts(self.bytes.as_ptr().cast::<u8>(), len);
            str::from_utf8_unchecked(slice)
        }
    }
}

/// A floating point number, f32 or f64, that can be written into a
/// [`zmij::Buffer`][Buffer].
///
/// This trait is sealed and cannot be implemented for types outside of the
/// `zmij` crate.
#[allow(unknown_lints)] // rustc older than 1.74
#[allow(private_bounds)]
pub trait Float: private::Sealed {}
impl Float for f32 {}
impl Float for f64 {}

mod private {
    pub trait Sealed: crate::traits::Float {
        fn is_nonfinite(self) -> bool;
        fn format_nonfinite(self) -> &'static str;
        unsafe fn write_to_zmij_buffer(self, buffer: *mut u8) -> *mut u8;
    }

    impl Sealed for f32 {
        #[inline]
        fn is_nonfinite(self) -> bool {
            const EXP_MASK: u32 = 0x7f800000;
            let bits = self.to_bits();
            bits & EXP_MASK == EXP_MASK
        }

        #[cold]
        #[cfg_attr(feature = "no-panic", inline)]
        fn format_nonfinite(self) -> &'static str {
            const MANTISSA_MASK: u32 = 0x007fffff;
            const SIGN_MASK: u32 = 0x80000000;
            let bits = self.to_bits();
            if bits & MANTISSA_MASK != 0 {
                crate::NAN
            } else if bits & SIGN_MASK != 0 {
                crate::NEG_INFINITY
            } else {
                crate::INFINITY
            }
        }

        #[cfg_attr(feature = "no-panic", inline)]
        unsafe fn write_to_zmij_buffer(self, buffer: *mut u8) -> *mut u8 {
            unsafe { crate::write(self, buffer) }
        }
    }

    impl Sealed for f64 {
        #[inline]
        fn is_nonfinite(self) -> bool {
            const EXP_MASK: u64 = 0x7ff0000000000000;
            let bits = self.to_bits();
            bits & EXP_MASK == EXP_MASK
        }

        #[cold]
        #[cfg_attr(feature = "no-panic", inline)]
        fn format_nonfinite(self) -> &'static str {
            const MANTISSA_MASK: u64 = 0x000fffffffffffff;
            const SIGN_MASK: u64 = 0x8000000000000000;
            let bits = self.to_bits();
            if bits & MANTISSA_MASK != 0 {
                crate::NAN
            } else if bits & SIGN_MASK != 0 {
                crate::NEG_INFINITY
            } else {
                crate::INFINITY
            }
        }

        #[cfg_attr(feature = "no-panic", inline)]
        unsafe fn write_to_zmij_buffer(self, buffer: *mut u8) -> *mut u8 {
            unsafe { crate::write(self, buffer) }
        }
    }
}

impl Default for Buffer {
    #[inline]
    #[cfg_attr(feature = "no-panic", no_panic)]
    fn default() -> Self {
        Buffer::new()
    }
}