memspan 0.1.0

SIMD-accelerated byte-class scanning for lexers and parsers. Backends: AVX-512, AVX2, SSE4.1, NEON, WASM SIMD128. no_std compatible.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200

#![allow(unsafe_op_in_unsafe_fn)]

#[cfg(target_arch = "x86")]
use core::arch::x86::*;
#[cfg(target_arch = "x86_64")]
use core::arch::x86_64::*;

// ── SSE2 / SSE4.2  (128-bit, __m128i) ───────────────────────────────────────

/// Returns an `__m128i` where each byte lane is `0xFF` if the corresponding
/// byte in `chunk` matches any needle, or `0x00` otherwise.
///
/// Iterates over needles with an OR-accumulator. Used for slices longer than
/// 8 elements where a const-unrolled tree is not available.
#[cfg_attr(not(tarpaulin), inline)]
#[target_feature(enable = "sse2")]
pub(in crate::needles) unsafe fn eq_any_mask_dynamic_sse2(
  chunk: __m128i,
  needles: &[u8],
) -> __m128i {
  let mut acc = _mm_setzero_si128();
  for &n in needles {
    acc = _mm_or_si128(acc, _mm_cmpeq_epi8(chunk, _mm_set1_epi8(n as i8)));
  }
  acc
}

/// Const-dispatch variant: unrolled balanced OR tree for 0–8 needles.
#[cfg_attr(not(tarpaulin), inline)]
#[target_feature(enable = "sse2")]
pub(in crate::needles) unsafe fn eq_any_mask_const_sse2<const N: usize>(
  chunk: __m128i,
  needles: [u8; N],
) -> __m128i {
  macro_rules! cmp {
    ($i:expr) => {
      _mm_cmpeq_epi8(chunk, _mm_set1_epi8(needles[$i] as i8))
    };
  }
  match N {
    0 => _mm_setzero_si128(),
    1 => cmp!(0),
    2 => _mm_or_si128(cmp!(0), cmp!(1)),
    3 => _mm_or_si128(_mm_or_si128(cmp!(0), cmp!(1)), cmp!(2)),
    4 => _mm_or_si128(
      _mm_or_si128(cmp!(0), cmp!(1)),
      _mm_or_si128(cmp!(2), cmp!(3)),
    ),
    5 => _mm_or_si128(
      _mm_or_si128(
        _mm_or_si128(cmp!(0), cmp!(1)),
        _mm_or_si128(cmp!(2), cmp!(3)),
      ),
      cmp!(4),
    ),
    6 => _mm_or_si128(
      _mm_or_si128(
        _mm_or_si128(cmp!(0), cmp!(1)),
        _mm_or_si128(cmp!(2), cmp!(3)),
      ),
      _mm_or_si128(cmp!(4), cmp!(5)),
    ),
    7 => _mm_or_si128(
      _mm_or_si128(
        _mm_or_si128(cmp!(0), cmp!(1)),
        _mm_or_si128(cmp!(2), cmp!(3)),
      ),
      _mm_or_si128(_mm_or_si128(cmp!(4), cmp!(5)), cmp!(6)),
    ),
    8 => _mm_or_si128(
      _mm_or_si128(
        _mm_or_si128(cmp!(0), cmp!(1)),
        _mm_or_si128(cmp!(2), cmp!(3)),
      ),
      _mm_or_si128(
        _mm_or_si128(cmp!(4), cmp!(5)),
        _mm_or_si128(cmp!(6), cmp!(7)),
      ),
    ),
    _ => eq_any_mask_dynamic_sse2(chunk, &needles),
  }
}

// ── AVX2  (256-bit, __m256i) ─────────────────────────────────────────────────

/// Returns a `__m256i` where each byte lane is `0xFF` if the byte in `chunk`
/// matches any needle.
#[cfg(target_arch = "x86_64")]
#[cfg_attr(not(tarpaulin), inline)]
#[target_feature(enable = "avx2")]
pub(in crate::needles) unsafe fn eq_any_mask_dynamic_avx2(
  chunk: __m256i,
  needles: &[u8],
) -> __m256i {
  let mut acc = _mm256_setzero_si256();
  for &n in needles {
    acc = _mm256_or_si256(acc, _mm256_cmpeq_epi8(chunk, _mm256_set1_epi8(n as i8)));
  }
  acc
}

/// Const-dispatch variant for AVX2: unrolled balanced OR tree for 0–8 needles.
#[cfg(target_arch = "x86_64")]
#[cfg_attr(not(tarpaulin), inline)]
#[target_feature(enable = "avx2")]
pub(in crate::needles) unsafe fn eq_any_mask_const_avx2<const N: usize>(
  chunk: __m256i,
  needles: [u8; N],
) -> __m256i {
  macro_rules! cmp {
    ($i:expr) => {
      _mm256_cmpeq_epi8(chunk, _mm256_set1_epi8(needles[$i] as i8))
    };
  }
  match N {
    0 => _mm256_setzero_si256(),
    1 => cmp!(0),
    2 => _mm256_or_si256(cmp!(0), cmp!(1)),
    3 => _mm256_or_si256(_mm256_or_si256(cmp!(0), cmp!(1)), cmp!(2)),
    4 => _mm256_or_si256(
      _mm256_or_si256(cmp!(0), cmp!(1)),
      _mm256_or_si256(cmp!(2), cmp!(3)),
    ),
    5 => _mm256_or_si256(
      _mm256_or_si256(
        _mm256_or_si256(cmp!(0), cmp!(1)),
        _mm256_or_si256(cmp!(2), cmp!(3)),
      ),
      cmp!(4),
    ),
    6 => _mm256_or_si256(
      _mm256_or_si256(
        _mm256_or_si256(cmp!(0), cmp!(1)),
        _mm256_or_si256(cmp!(2), cmp!(3)),
      ),
      _mm256_or_si256(cmp!(4), cmp!(5)),
    ),
    7 => _mm256_or_si256(
      _mm256_or_si256(
        _mm256_or_si256(cmp!(0), cmp!(1)),
        _mm256_or_si256(cmp!(2), cmp!(3)),
      ),
      _mm256_or_si256(_mm256_or_si256(cmp!(4), cmp!(5)), cmp!(6)),
    ),
    8 => _mm256_or_si256(
      _mm256_or_si256(
        _mm256_or_si256(cmp!(0), cmp!(1)),
        _mm256_or_si256(cmp!(2), cmp!(3)),
      ),
      _mm256_or_si256(
        _mm256_or_si256(cmp!(4), cmp!(5)),
        _mm256_or_si256(cmp!(6), cmp!(7)),
      ),
    ),
    _ => eq_any_mask_dynamic_avx2(chunk, &needles),
  }
}

// ── AVX-512BW  (512-bit, __m512i → u64 mask) ─────────────────────────────────

/// Returns a `u64` bitmask where bit `i` is set if byte lane `i` of `chunk`
/// matches any needle. Uses `_mm512_cmpeq_epi8_mask` which returns the mask
/// directly — no `movemask` conversion needed.
#[cfg(target_arch = "x86_64")]
#[cfg_attr(not(tarpaulin), inline)]
#[target_feature(enable = "avx512bw")]
pub(in crate::needles) unsafe fn eq_any_mask_dynamic_avx512(chunk: __m512i, needles: &[u8]) -> u64 {
  let mut acc: u64 = 0;
  for &n in needles {
    acc |= _mm512_cmpeq_epi8_mask(chunk, _mm512_set1_epi8(n as i8));
  }
  acc
}

/// Const-dispatch variant for AVX-512BW: unrolled balanced OR for 0–8 needles.
#[cfg(target_arch = "x86_64")]
#[cfg_attr(not(tarpaulin), inline)]
#[target_feature(enable = "avx512bw")]
pub(in crate::needles) unsafe fn eq_any_mask_const_avx512<const N: usize>(
  chunk: __m512i,
  needles: [u8; N],
) -> u64 {
  macro_rules! cmp {
    ($i:expr) => {
      _mm512_cmpeq_epi8_mask(chunk, _mm512_set1_epi8(needles[$i] as i8))
    };
  }
  match N {
    0 => 0u64,
    1 => cmp!(0),
    2 => cmp!(0) | cmp!(1),
    3 => cmp!(0) | cmp!(1) | cmp!(2),
    4 => (cmp!(0) | cmp!(1)) | (cmp!(2) | cmp!(3)),
    5 => (cmp!(0) | cmp!(1) | cmp!(2) | cmp!(3)) | cmp!(4),
    6 => (cmp!(0) | cmp!(1) | cmp!(2) | cmp!(3)) | (cmp!(4) | cmp!(5)),
    7 => (cmp!(0) | cmp!(1) | cmp!(2) | cmp!(3)) | (cmp!(4) | cmp!(5) | cmp!(6)),
    8 => (cmp!(0) | cmp!(1) | cmp!(2) | cmp!(3)) | (cmp!(4) | cmp!(5) | cmp!(6) | cmp!(7)),
    _ => eq_any_mask_dynamic_avx512(chunk, &needles),
  }
}