|
| 1 | +//! Common utilities, for internal use only. |
| 2 | +
|
| 3 | +use crate::ptr; |
| 4 | + |
| 5 | +/// Helper methods to process immutable bytes. |
| 6 | +pub(crate) trait ByteSlice: AsRef<[u8]> { |
| 7 | + unsafe fn first_unchecked(&self) -> u8 { |
| 8 | + debug_assert!(!self.is_empty()); |
| 9 | + // SAFETY: safe as long as self is not empty |
| 10 | + unsafe { *self.as_ref().get_unchecked(0) } |
| 11 | + } |
| 12 | + |
| 13 | + /// Get if the slice contains no elements. |
| 14 | + fn is_empty(&self) -> bool { |
| 15 | + self.as_ref().is_empty() |
| 16 | + } |
| 17 | + |
| 18 | + /// Check if the slice at least `n` length. |
| 19 | + fn check_len(&self, n: usize) -> bool { |
| 20 | + n <= self.as_ref().len() |
| 21 | + } |
| 22 | + |
| 23 | + /// Check if the first character in the slice is equal to c. |
| 24 | + fn first_is(&self, c: u8) -> bool { |
| 25 | + self.as_ref().first() == Some(&c) |
| 26 | + } |
| 27 | + |
| 28 | + /// Check if the first character in the slice is equal to c1 or c2. |
| 29 | + fn first_is2(&self, c1: u8, c2: u8) -> bool { |
| 30 | + if let Some(&c) = self.as_ref().first() { c == c1 || c == c2 } else { false } |
| 31 | + } |
| 32 | + |
| 33 | + /// Bounds-checked test if the first character in the slice is a digit. |
| 34 | + fn first_isdigit(&self) -> bool { |
| 35 | + if let Some(&c) = self.as_ref().first() { c.is_ascii_digit() } else { false } |
| 36 | + } |
| 37 | + |
| 38 | + /// Check if self starts with u with a case-insensitive comparison. |
| 39 | + fn eq_ignore_case(&self, u: &[u8]) -> bool { |
| 40 | + debug_assert!(self.as_ref().len() >= u.len()); |
| 41 | + let iter = self.as_ref().iter().zip(u.iter()); |
| 42 | + let d = iter.fold(0, |i, (&x, &y)| i | (x ^ y)); |
| 43 | + d == 0 || d == 32 |
| 44 | + } |
| 45 | + |
| 46 | + /// Get the remaining slice after the first N elements. |
| 47 | + fn advance(&self, n: usize) -> &[u8] { |
| 48 | + &self.as_ref()[n..] |
| 49 | + } |
| 50 | + |
| 51 | + /// Get the slice after skipping all leading characters equal c. |
| 52 | + fn skip_chars(&self, c: u8) -> &[u8] { |
| 53 | + let mut s = self.as_ref(); |
| 54 | + while s.first_is(c) { |
| 55 | + s = s.advance(1); |
| 56 | + } |
| 57 | + s |
| 58 | + } |
| 59 | + |
| 60 | + /// Get the slice after skipping all leading characters equal c1 or c2. |
| 61 | + fn skip_chars2(&self, c1: u8, c2: u8) -> &[u8] { |
| 62 | + let mut s = self.as_ref(); |
| 63 | + while s.first_is2(c1, c2) { |
| 64 | + s = s.advance(1); |
| 65 | + } |
| 66 | + s |
| 67 | + } |
| 68 | + |
| 69 | + /// Read 8 bytes as a 64-bit integer in little-endian order. |
| 70 | + unsafe fn read_u64_unchecked(&self) -> u64 { |
| 71 | + debug_assert!(self.check_len(8)); |
| 72 | + let src = self.as_ref().as_ptr() as *const u64; |
| 73 | + // SAFETY: safe as long as self is at least 8 bytes |
| 74 | + u64::from_le(unsafe { ptr::read_unaligned(src) }) |
| 75 | + } |
| 76 | + |
| 77 | + /// Try to read the next 8 bytes from the slice. |
| 78 | + fn read_u64(&self) -> Option<u64> { |
| 79 | + if self.check_len(8) { |
| 80 | + // SAFETY: self must be at least 8 bytes. |
| 81 | + Some(unsafe { self.read_u64_unchecked() }) |
| 82 | + } else { |
| 83 | + None |
| 84 | + } |
| 85 | + } |
| 86 | + |
| 87 | + /// Calculate the offset of slice from another. |
| 88 | + fn offset_from(&self, other: &Self) -> isize { |
| 89 | + other.as_ref().len() as isize - self.as_ref().len() as isize |
| 90 | + } |
| 91 | +} |
| 92 | + |
| 93 | +impl ByteSlice for [u8] {} |
| 94 | + |
| 95 | +/// Helper methods to process mutable bytes. |
| 96 | +pub(crate) trait ByteSliceMut: AsMut<[u8]> { |
| 97 | + /// Write a 64-bit integer as 8 bytes in little-endian order. |
| 98 | + unsafe fn write_u64_unchecked(&mut self, value: u64) { |
| 99 | + debug_assert!(self.as_mut().len() >= 8); |
| 100 | + let dst = self.as_mut().as_mut_ptr() as *mut u64; |
| 101 | + // NOTE: we must use `write_unaligned`, since dst is not |
| 102 | + // guaranteed to be properly aligned. Miri will warn us |
| 103 | + // if we use `write` instead of `write_unaligned`, as expected. |
| 104 | + // SAFETY: safe as long as self is at least 8 bytes |
| 105 | + unsafe { |
| 106 | + ptr::write_unaligned(dst, u64::to_le(value)); |
| 107 | + } |
| 108 | + } |
| 109 | +} |
| 110 | + |
| 111 | +impl ByteSliceMut for [u8] {} |
| 112 | + |
| 113 | +/// Bytes wrapper with specialized methods for ASCII characters. |
| 114 | +#[derive(Debug, Clone, Copy, PartialEq, Eq)] |
| 115 | +pub(crate) struct AsciiStr<'a> { |
| 116 | + slc: &'a [u8], |
| 117 | +} |
| 118 | + |
| 119 | +impl<'a> AsciiStr<'a> { |
| 120 | + pub fn new(slc: &'a [u8]) -> Self { |
| 121 | + Self { slc } |
| 122 | + } |
| 123 | + |
| 124 | + /// Advance the view by n, advancing it in-place to (n..). |
| 125 | + pub unsafe fn step_by(&mut self, n: usize) -> &mut Self { |
| 126 | + // SAFETY: safe as long n is less than the buffer length |
| 127 | + self.slc = unsafe { self.slc.get_unchecked(n..) }; |
| 128 | + self |
| 129 | + } |
| 130 | + |
| 131 | + /// Advance the view by n, advancing it in-place to (1..). |
| 132 | + pub unsafe fn step(&mut self) -> &mut Self { |
| 133 | + // SAFETY: safe as long as self is not empty |
| 134 | + unsafe { self.step_by(1) } |
| 135 | + } |
| 136 | + |
| 137 | + /// Iteratively parse and consume digits from bytes. |
| 138 | + pub fn parse_digits(&mut self, mut func: impl FnMut(u8)) { |
| 139 | + while let Some(&c) = self.as_ref().first() { |
| 140 | + let c = c.wrapping_sub(b'0'); |
| 141 | + if c < 10 { |
| 142 | + func(c); |
| 143 | + // SAFETY: self cannot be empty |
| 144 | + unsafe { |
| 145 | + self.step(); |
| 146 | + } |
| 147 | + } else { |
| 148 | + break; |
| 149 | + } |
| 150 | + } |
| 151 | + } |
| 152 | +} |
| 153 | + |
| 154 | +impl<'a> AsRef<[u8]> for AsciiStr<'a> { |
| 155 | + #[inline] |
| 156 | + fn as_ref(&self) -> &[u8] { |
| 157 | + self.slc |
| 158 | + } |
| 159 | +} |
| 160 | + |
| 161 | +impl<'a> ByteSlice for AsciiStr<'a> {} |
| 162 | + |
| 163 | +/// Determine if 8 bytes are all decimal digits. |
| 164 | +/// This does not care about the order in which the bytes were loaded. |
| 165 | +pub(crate) fn is_8digits(v: u64) -> bool { |
| 166 | + let a = v.wrapping_add(0x4646_4646_4646_4646); |
| 167 | + let b = v.wrapping_sub(0x3030_3030_3030_3030); |
| 168 | + (a | b) & 0x8080_8080_8080_8080 == 0 |
| 169 | +} |
| 170 | + |
| 171 | +/// Iteratively parse and consume digits from bytes. |
| 172 | +pub(crate) fn parse_digits(s: &mut &[u8], mut f: impl FnMut(u8)) { |
| 173 | + while let Some(&c) = s.get(0) { |
| 174 | + let c = c.wrapping_sub(b'0'); |
| 175 | + if c < 10 { |
| 176 | + f(c); |
| 177 | + *s = s.advance(1); |
| 178 | + } else { |
| 179 | + break; |
| 180 | + } |
| 181 | + } |
| 182 | +} |
| 183 | + |
| 184 | +/// A custom 64-bit floating point type, representing `f * 2^e`. |
| 185 | +/// e is biased, so it be directly shifted into the exponent bits. |
| 186 | +#[derive(Debug, Copy, Clone, PartialEq, Eq, Default)] |
| 187 | +pub struct BiasedFp { |
| 188 | + /// The significant digits. |
| 189 | + pub f: u64, |
| 190 | + /// The biased, binary exponent. |
| 191 | + pub e: i32, |
| 192 | +} |
| 193 | + |
| 194 | +impl BiasedFp { |
| 195 | + pub const fn zero_pow2(e: i32) -> Self { |
| 196 | + Self { f: 0, e } |
| 197 | + } |
| 198 | +} |
0 commit comments