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
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
use criterion::{BenchmarkId, Criterion, Throughput, criterion_group, criterion_main};
use memspan::{Needles, skip};
use std::hint::black_box;

/// Sizes for the per-call micro benches. Includes points around the SIMD
/// dispatch threshold (`NEON_CHUNK_SIZE = 16`) so the boundary cost is visible.
const MICRO_LENGTHS: [usize; 8] = [15, 16, 17, 32, 64, 256, 4 * 1024, 64 * 1024];

/// Sizes for the workload benches. Two pages-ish targets, one well past LLC.
const WORKLOAD_LENGTHS: [usize; 2] = [8 * 1024, 128 * 1024];

/// Gap between consecutive needle hits used by the density-sweep bench. Smaller
/// values flip the SIMD-vs-scalar verdict in favor of the scalar path because
/// the per-call constant dominates.
const DENSITY_GAPS: [usize; 5] = [4, 16, 64, 256, 1024];

fn input_with_match_at_end(len: usize, fill: u8, needle: u8) -> Vec<u8> {
  let mut input = vec![fill; len];

  if let Some(last) = input.last_mut() {
    *last = needle;
  }

  input
}

/// Build a `len`-byte buffer where `needle` appears at every `gap` bytes.
fn input_with_periodic_match(len: usize, fill: u8, needle: u8, gap: usize) -> Vec<u8> {
  assert!(gap >= 1, "gap must be at least 1");
  let mut input = vec![fill; len];
  let mut pos = gap.saturating_sub(1);
  while pos < len {
    input[pos] = needle;
    pos += gap;
  }
  input
}

fn repeat_to_len(fragment: &[u8], target_len: usize) -> Vec<u8> {
  let mut input = Vec::with_capacity(target_len + fragment.len());

  while input.len() < target_len {
    input.extend_from_slice(fragment);
  }

  input.truncate(target_len);
  input
}

fn scalar_tail_find<Nd>(input: &[u8], needles: Nd) -> Option<usize>
where
  Nd: Needles,
{
  if needles.is_empty() {
    return None;
  }

  needles.tail_find(input)
}

fn scan_all_dispatch<Nd>(input: &[u8], needles: Nd) -> usize
where
  Nd: Needles + Copy,
{
  let mut pos = 0;
  let mut checksum = 0usize;

  while pos < input.len() {
    let Some(hit) = skip::skip_until(&input[pos..], needles) else {
      break;
    };

    pos += hit;
    checksum = checksum.wrapping_add(pos);
    pos += 1;
  }

  checksum
}

fn scan_all_scalar<Nd>(input: &[u8], needles: Nd) -> usize
where
  Nd: Needles + Copy,
{
  let mut pos = 0;
  let mut checksum = 0usize;

  while pos < input.len() {
    let Some(hit) = scalar_tail_find(&input[pos..], needles) else {
      break;
    };

    pos += hit;
    checksum = checksum.wrapping_add(pos);
    pos += 1;
  }

  checksum
}

/// Per-call micro: full-length scan, only one match at the very end. Measures
/// raw throughput of a single `skip_until` call across a range of sizes that
/// straddle the SIMD dispatch threshold.
fn bench_micro_end_match(c: &mut Criterion) {
  const NEEDLES: [u8; 5] = [b' ', b'\t', b'\r', b'\n', b','];

  let mut group = c.benchmark_group("skip_until/micro/end_match");

  for len in MICRO_LENGTHS {
    let input = input_with_match_at_end(len, b'a', b'\n');
    group.throughput(Throughput::Bytes(len as u64));

    group.bench_with_input(
      BenchmarkId::new("dispatch_fixed", len),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(skip::skip_until(
            black_box(input.as_slice()),
            black_box(NEEDLES),
          ))
        })
      },
    );

    group.bench_with_input(
      BenchmarkId::new("scalar_tail_find", len),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(scalar_tail_find(
            black_box(input.as_slice()),
            black_box(NEEDLES),
          ))
        })
      },
    );
  }

  group.finish();
}

/// Density sweep on a fixed buffer: same input length, varying gap between
/// needle hits. This isolates the variable that flips the SIMD-vs-scalar
/// verdict and exposes where the crossover lives.
fn bench_density_sweep(c: &mut Criterion) {
  const NEEDLES: [u8; 5] = [b' ', b'\t', b'\r', b'\n', b','];
  const LEN: usize = 64 * 1024;

  let mut group = c.benchmark_group("skip_until/density_sweep");

  for gap in DENSITY_GAPS {
    let input = input_with_periodic_match(LEN, b'a', b'\n', gap);
    group.throughput(Throughput::Bytes(LEN as u64));

    group.bench_with_input(
      BenchmarkId::new("dispatch_fixed", gap),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(scan_all_dispatch(
            black_box(input.as_slice()),
            black_box(NEEDLES),
          ))
        })
      },
    );

    group.bench_with_input(
      BenchmarkId::new("scalar_tail_find", gap),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(scan_all_scalar(
            black_box(input.as_slice()),
            black_box(NEEDLES),
          ))
        })
      },
    );
  }

  group.finish();
}

fn bench_graphql_ignored_workload(c: &mut Criterion) {
  const NEEDLES: [u8; 5] = [b' ', b'\t', b'\r', b'\n', b','];
  const FRAGMENT: &[u8] = br#"
query GetUser($id: ID!, $limit: Int = 10) {
  user(id: $id) {
    id
    name
    friends(first: $limit, after: null) {
      nodes {
        id
        name
        profile { avatarUrl(size: 64), bio }
      }
    }
  }
}

fragment UserFields on User {
  id
  name
  email
}
"#;

  let mut group = c.benchmark_group("skip_until/workload/graphql_ignored_scan_all");

  for len in WORKLOAD_LENGTHS {
    let input = repeat_to_len(FRAGMENT, len);
    let dynamic_needles: &[u8] = &NEEDLES;
    group.throughput(Throughput::Bytes(len as u64));

    group.bench_with_input(
      BenchmarkId::new("dispatch_fixed", len),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(scan_all_dispatch(
            black_box(input.as_slice()),
            black_box(NEEDLES),
          ))
        })
      },
    );

    group.bench_with_input(
      BenchmarkId::new("dispatch_dynamic", len),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(scan_all_dispatch(
            black_box(input.as_slice()),
            black_box(dynamic_needles),
          ))
        })
      },
    );

    group.bench_with_input(
      BenchmarkId::new("scalar_tail_find", len),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(scan_all_scalar(
            black_box(input.as_slice()),
            black_box(NEEDLES),
          ))
        })
      },
    );
  }

  group.finish();
}

fn bench_quote_or_escape_workload(c: &mut Criterion) {
  const NEEDLES: [u8; 2] = [b'"', b'\\'];
  const FRAGMENT: &[u8] =
    br#"name: "Ada Lovelace", bio: "first programmer \"notes\" \\ archive", city: "London"
"#;

  let mut group = c.benchmark_group("skip_until/workload/quote_or_escape_scan_all");

  for len in WORKLOAD_LENGTHS {
    let input = repeat_to_len(FRAGMENT, len);
    let dynamic_needles: &[u8] = &NEEDLES;
    group.throughput(Throughput::Bytes(len as u64));

    group.bench_with_input(
      BenchmarkId::new("dispatch_fixed", len),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(scan_all_dispatch(
            black_box(input.as_slice()),
            black_box(NEEDLES),
          ))
        })
      },
    );

    group.bench_with_input(
      BenchmarkId::new("dispatch_dynamic", len),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(scan_all_dispatch(
            black_box(input.as_slice()),
            black_box(dynamic_needles),
          ))
        })
      },
    );

    group.bench_with_input(
      BenchmarkId::new("scalar_tail_find", len),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(scan_all_scalar(
            black_box(input.as_slice()),
            black_box(NEEDLES),
          ))
        })
      },
    );

    // Direct memchr2 baseline: bypasses the Needles trait entirely so any
    // wrapper overhead shows up as the gap between this row and
    // `dispatch_fixed`.
    group.bench_with_input(
      BenchmarkId::new("memchr2_direct", len),
      &input,
      |b, input| {
        b.iter(|| {
          let buf: &[u8] = black_box(input.as_slice());
          let mut pos = 0;
          let mut checksum = 0usize;
          while pos < buf.len() {
            let Some(hit) = memchr::memchr2(b'"', b'\\', &buf[pos..]) else {
              break;
            };
            pos += hit;
            checksum = checksum.wrapping_add(pos);
            pos += 1;
          }
          black_box(checksum)
        })
      },
    );
  }

  group.finish();
}

fn bench_comment_newline_workload(c: &mut Criterion) {
  const FRAGMENT: &[u8] = br#"# warm cache comment with enough text to scan before newline
# another GraphQL comment that ends at the next line terminator
# short
"#;

  let mut group = c.benchmark_group("skip_until/workload/comment_newline_scan_all");

  for len in WORKLOAD_LENGTHS {
    let input = repeat_to_len(FRAGMENT, len);
    let dynamic_needle: &[u8] = b"\n";
    group.throughput(Throughput::Bytes(len as u64));

    group.bench_with_input(BenchmarkId::new("dispatch_u8", len), &input, |b, input| {
      b.iter(|| {
        black_box(scan_all_dispatch(
          black_box(input.as_slice()),
          black_box(b'\n'),
        ))
      })
    });

    group.bench_with_input(
      BenchmarkId::new("dispatch_dynamic", len),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(scan_all_dispatch(
            black_box(input.as_slice()),
            black_box(dynamic_needle),
          ))
        })
      },
    );

    group.bench_with_input(
      BenchmarkId::new("scalar_tail_find", len),
      &input,
      |b, input| {
        b.iter(|| {
          black_box(scan_all_scalar(
            black_box(input.as_slice()),
            black_box(b'\n'),
          ))
        })
      },
    );

    // Direct memchr baseline.
    group.bench_with_input(
      BenchmarkId::new("memchr_direct", len),
      &input,
      |b, input| {
        b.iter(|| {
          let buf: &[u8] = black_box(input.as_slice());
          let mut pos = 0;
          let mut checksum = 0usize;
          while pos < buf.len() {
            let Some(hit) = memchr::memchr(b'\n', &buf[pos..]) else {
              break;
            };
            pos += hit;
            checksum = checksum.wrapping_add(pos);
            pos += 1;
          }
          black_box(checksum)
        })
      },
    );
  }

  group.finish();
}

criterion_group!(
  benches,
  bench_micro_end_match,
  bench_density_sweep,
  bench_graphql_ignored_workload,
  bench_quote_or_escape_workload,
  bench_comment_newline_workload
);
criterion_main!(benches);