wasm-bitpack 0.1.0

SIMD-accelerated integer bit-packing with a real WebAssembly SIMD128 decode fast path
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
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
//! [`Scalar`] — the pure safe-Rust reference bit-packing implementation, always available.

use crate::traits::BitPacker;

/// Portable, safe-Rust reference implementation of [`BitPacker`].
///
/// Correct on every target, including under miri. Every other implementation in this
/// crate is checked for byte-for-byte equivalence against `Scalar`'s output.
///
/// # Examples
///
/// ```
/// use wasm_bitpack::{BitPacker, Scalar};
///
/// let values = vec![7u32; Scalar::BLOCK_LEN];
/// let num_bits = Scalar::num_bits(&values);
///
/// let mut compressed = vec![0u8; (Scalar::BLOCK_LEN * num_bits as usize + 7) / 8];
/// let written = Scalar::compress(&values, &mut compressed, num_bits);
///
/// let mut decompressed = vec![0u32; Scalar::BLOCK_LEN];
/// Scalar::decompress(&compressed[..written], &mut decompressed, num_bits);
/// assert_eq!(decompressed, values);
/// ```
pub struct Scalar;

impl BitPacker for Scalar {
    const BLOCK_LEN: usize = 128;

    fn num_bits(decompressed: &[u32]) -> u8 {
        let max = decompressed.iter().copied().max().unwrap_or(0);
        bits_for_max_value(max)
    }

    fn compress(decompressed: &[u32], compressed: &mut [u8], num_bits: u8) -> usize {
        assert_eq!(
            decompressed.len(),
            Self::BLOCK_LEN,
            "decompressed.len() must be exactly BLOCK_LEN ({}), got {}",
            Self::BLOCK_LEN,
            decompressed.len()
        );
        check_num_bits(num_bits);
        let needed = packed_len_bytes(Self::BLOCK_LEN, num_bits);
        assert!(
            compressed.len() >= needed,
            "compressed buffer too small: need at least {needed} bytes, got {}",
            compressed.len()
        );

        bitpack_interleaved_into(decompressed, compressed, num_bits)
    }

    fn decompress(compressed: &[u8], decompressed: &mut [u32], num_bits: u8) -> usize {
        assert_eq!(
            decompressed.len(),
            Self::BLOCK_LEN,
            "decompressed.len() must be exactly BLOCK_LEN ({}), got {}",
            Self::BLOCK_LEN,
            decompressed.len()
        );
        check_num_bits(num_bits);
        let needed = packed_len_bytes(Self::BLOCK_LEN, num_bits);
        assert!(
            compressed.len() >= needed,
            "compressed buffer too small: need at least {needed} bytes, got {}",
            compressed.len()
        );

        bitunpack_interleaved_into(compressed, decompressed, num_bits)
    }
}

/// The minimum number of bits needed to represent `max` (`0` for `max == 0`).
///
/// `pub(crate)` (not private) so [`crate::sorted`]'s `num_bits_sorted`/
/// `num_bits_strictly_sorted` can reuse it for their own max-delta-to-bit-width conversion,
/// instead of a second copy of the same `leading_zeros` trick.
#[allow(clippy::cast_possible_truncation)] // `32 - leading_zeros()` is always in 0..=32
pub(crate) fn bits_for_max_value(max: u32) -> u8 {
    (32 - max.leading_zeros()) as u8
}

/// Panics unless `num_bits` is in the valid `1..=32` range.
///
/// Shared between [`Scalar`], and the [`pack`](crate::pack)/[`unpack`](crate::unpack)
/// wrapper, so every entry point rejects an out-of-range bit width with the same message.
pub(crate) fn check_num_bits(num_bits: u8) {
    assert!(
        (1..=32).contains(&num_bits),
        "num_bits must be in 1..=32, got {num_bits}"
    );
}

/// The number of bytes needed to tightly bit-pack `len` values at `num_bits` bits each.
pub(crate) fn packed_len_bytes(len: usize, num_bits: u8) -> usize {
    let total_bits = len as u64 * u64::from(num_bits);
    ((total_bits + 7) / 8) as usize
}

/// Tightly bit-packs `values` into `out` at `num_bits` bits per value (LSB-first: value
/// `i`'s bit `0` is the `(i * num_bits)`-th bit of the output stream, counting from the
/// least-significant bit of `out[0]`), returning the number of bytes written.
///
/// No length restriction — this is the primitive both [`Scalar::compress`] (exact
/// `BLOCK_LEN` blocks) and the [`pack`](crate::pack) wrapper's partial-block tail build on.
///
/// Trusts its caller: `out` must already be verified to be at least
/// `packed_len_bytes(values.len(), num_bits)` bytes, and `num_bits` must already be
/// verified to be in `1..=32`.
#[allow(clippy::cast_possible_truncation)] // `acc & 0xff` always fits in a u8
pub(crate) fn bitpack_into(values: &[u32], out: &mut [u8], num_bits: u8) -> usize {
    let mask = (1u64 << num_bits) - 1;
    let mut acc: u64 = 0;
    let mut acc_bits: u32 = 0;
    let mut out_pos = 0usize;

    for &v in values {
        acc |= (u64::from(v) & mask) << acc_bits;
        acc_bits += u32::from(num_bits);
        while acc_bits >= 8 {
            out[out_pos] = (acc & 0xff) as u8;
            out_pos += 1;
            acc >>= 8;
            acc_bits -= 8;
        }
    }
    if acc_bits > 0 {
        out[out_pos] = (acc & 0xff) as u8;
        out_pos += 1;
    }

    out_pos
}

/// Unpacks `out.len()` values from `bytes`, the inverse of [`bitpack_into`], returning the
/// number of bytes read.
///
/// Trusts its caller in the same way [`bitpack_into`] does.
#[allow(clippy::cast_possible_truncation)] // `acc & mask` fits in a u32 for num_bits <= 32
pub(crate) fn bitunpack_into(bytes: &[u8], out: &mut [u32], num_bits: u8) -> usize {
    let mask = (1u64 << num_bits) - 1;
    let mut acc: u64 = 0;
    let mut acc_bits: u32 = 0;
    let mut in_pos = 0usize;

    for slot in out.iter_mut() {
        while acc_bits < u32::from(num_bits) {
            acc |= u64::from(bytes[in_pos]) << acc_bits;
            in_pos += 1;
            acc_bits += 8;
        }
        *slot = (acc & mask) as u32;
        acc >>= u32::from(num_bits);
        acc_bits -= u32::from(num_bits);
    }

    in_pos
}

/// Number of interleaved lanes the block-exact packed format below splits a block into — one
/// lane per `u32` position in the WASM SIMD128 register [`Wasm128`](crate::Wasm128)'s decode
/// kernel loads. See `plans/decisions/0010-bp128-style-packed-format.md` for the full
/// rationale.
const LANES: usize = 4;

/// Tightly bit-packs a `BLOCK_LEN`-exact (128-value) block into `out` using the 4-way
/// interleaved layout [`Wasm128`](crate::Wasm128)'s decode kernel expects, returning the
/// number of bytes written: split `values` into 4 stride-4 sub-streams (lane `j` holds values
/// `j, j+4, ..., j+124`), bit-pack each lane independently via [`bitpack_into`], then
/// interleave the 4 lanes' packed 32-bit words round-robin (`out`'s first 4 bytes are lane 0's
/// first word, the next 4 are lane 1's first word, and so on). See
/// `plans/decisions/0010-bp128-style-packed-format.md`.
///
/// Because `values.len() / LANES` is always `32` (the only block length this crate supports),
/// each lane's packed length is always an exact multiple of 4 bytes for every `num_bits` in
/// `1..=32` (`32 * num_bits` bits is always a whole number of 32-bit words) — no padding, no
/// partial words, ever, and the total output size always equals
/// `packed_len_bytes(values.len(), num_bits)`, unchanged from the flat format.
///
/// Trusts its caller the same way [`bitpack_into`] does: `values.len()` must already be
/// exactly `BLOCK_LEN` (128), `out` at least `packed_len_bytes(values.len(), num_bits)` bytes,
/// and `num_bits` already verified to be in `1..=32`.
pub(crate) fn bitpack_interleaved_into(values: &[u32], out: &mut [u8], num_bits: u8) -> usize {
    let lane_len = values.len() / LANES;
    let lane_bytes = packed_len_bytes(lane_len, num_bits);

    let mut lane_values = [0u32; 32];
    let mut lane_packed = [0u8; 128];

    for lane in 0..LANES {
        for (i, slot) in lane_values[..lane_len].iter_mut().enumerate() {
            *slot = values[lane + i * LANES];
        }
        bitpack_into(
            &lane_values[..lane_len],
            &mut lane_packed[..lane_bytes],
            num_bits,
        );

        for (word, chunk) in lane_packed[..lane_bytes].chunks_exact(4).enumerate() {
            let dst = word * LANES * 4 + lane * 4;
            out[dst..dst + 4].copy_from_slice(chunk);
        }
    }

    lane_bytes * LANES
}

/// Unpacks `out.len()` values from `bytes`, the inverse of [`bitpack_interleaved_into`],
/// returning the number of bytes read.
///
/// Trusts its caller in the same way [`bitpack_interleaved_into`] does.
pub(crate) fn bitunpack_interleaved_into(bytes: &[u8], out: &mut [u32], num_bits: u8) -> usize {
    let lane_len = out.len() / LANES;
    let lane_bytes = packed_len_bytes(lane_len, num_bits);

    let mut lane_values = [0u32; 32];
    let mut lane_packed = [0u8; 128];

    for lane in 0..LANES {
        for (word, chunk) in lane_packed[..lane_bytes].chunks_exact_mut(4).enumerate() {
            let src = word * LANES * 4 + lane * 4;
            chunk.copy_from_slice(&bytes[src..src + 4]);
        }
        bitunpack_into(
            &lane_packed[..lane_bytes],
            &mut lane_values[..lane_len],
            num_bits,
        );

        for (i, &value) in lane_values[..lane_len].iter().enumerate() {
            out[lane + i * LANES] = value;
        }
    }

    lane_bytes * LANES
}

#[cfg(test)]
#[allow(clippy::cast_possible_truncation)] // test data lengths/values are always small
mod tests {
    use super::*;
    use crate::testing::{gen, oracle};
    use proptest::prelude::*;

    fn block(value: u32) -> Vec<u32> {
        vec![value; Scalar::BLOCK_LEN]
    }

    #[test]
    fn num_bits_all_zero_is_zero() {
        assert_eq!(Scalar::num_bits(&[0, 0, 0]), 0);
    }

    #[test]
    fn num_bits_matches_expected_widths() {
        assert_eq!(Scalar::num_bits(&[1]), 1);
        assert_eq!(Scalar::num_bits(&[2]), 2);
        assert_eq!(Scalar::num_bits(&[3]), 2);
        assert_eq!(Scalar::num_bits(&[u32::MAX]), 32);
    }

    #[test]
    fn round_trip_num_bits_1() {
        let values: Vec<u32> = (0..Scalar::BLOCK_LEN as u32).map(|i| i % 2).collect();
        let mut compressed = vec![0u8; packed_len_bytes(Scalar::BLOCK_LEN, 1)];
        let written = Scalar::compress(&values, &mut compressed, 1);
        assert_eq!(written, compressed.len());

        let mut decompressed = vec![0u32; Scalar::BLOCK_LEN];
        let read = Scalar::decompress(&compressed, &mut decompressed, 1);
        assert_eq!(read, compressed.len());
        assert_eq!(decompressed, values);
    }

    #[test]
    fn round_trip_num_bits_32() {
        let values: Vec<u32> = (0..Scalar::BLOCK_LEN as u32)
            .map(|i| u32::MAX - i)
            .collect();
        let mut compressed = vec![0u8; packed_len_bytes(Scalar::BLOCK_LEN, 32)];
        let written = Scalar::compress(&values, &mut compressed, 32);
        assert_eq!(written, compressed.len());

        let mut decompressed = vec![0u32; Scalar::BLOCK_LEN];
        Scalar::decompress(&compressed, &mut decompressed, 32);
        assert_eq!(decompressed, values);
    }

    #[test]
    fn compress_matches_oracle() {
        for num_bits in [1u8, 3, 7, 8, 11, 16, 17, 24, 32] {
            let modulus = 1u64 << num_bits;
            let values: Vec<u32> = (0..Scalar::BLOCK_LEN as u64)
                .map(|i| (i % modulus) as u32)
                .collect();

            let mut compressed = vec![0u8; packed_len_bytes(Scalar::BLOCK_LEN, num_bits)];
            Scalar::compress(&values, &mut compressed, num_bits);

            let expected = oracle::oracle_encode_interleaved(&values, num_bits);
            assert_eq!(compressed, expected, "mismatch at num_bits={num_bits}");
        }
    }

    #[test]
    fn interleaved_round_trip() {
        for num_bits in [1u8, 3, 7, 8, 11, 16, 17, 24, 32] {
            let modulus = 1u64 << num_bits;
            let values: Vec<u32> = (0..Scalar::BLOCK_LEN as u64)
                .map(|i| (i * 7 % modulus) as u32)
                .collect();

            let mut packed = vec![0u8; packed_len_bytes(Scalar::BLOCK_LEN, num_bits)];
            let written = bitpack_interleaved_into(&values, &mut packed, num_bits);
            assert_eq!(written, packed.len());

            let mut decoded = vec![0u32; Scalar::BLOCK_LEN];
            let read = bitunpack_interleaved_into(&packed, &mut decoded, num_bits);
            assert_eq!(read, packed.len());
            assert_eq!(
                decoded, values,
                "round-trip mismatch at num_bits={num_bits}"
            );
        }
    }

    /// Hand-verifiable layout check: at `num_bits == 32` each lane's `w`-th packed word is
    /// simply that lane's `w`-th value's raw little-endian bytes (no bit-level packing
    /// overlap), so the interleaved output at word `w`, lane `j` must be the raw bytes of
    /// `values[j + w * 4]` — e.g. word 0 holds values `0, 1, 2, 3` (one per lane) and word 1
    /// holds values `4, 5, 6, 7`.
    #[test]
    fn interleaved_known_layout_num_bits_32() {
        let values: Vec<u32> = (0..Scalar::BLOCK_LEN as u32).collect();
        let mut packed = vec![0u8; packed_len_bytes(Scalar::BLOCK_LEN, 32)];
        bitpack_interleaved_into(&values, &mut packed, 32);

        for word in 0..32usize {
            for lane in 0..4usize {
                let expected = (lane as u32 + word as u32 * 4).to_le_bytes();
                let offset = word * 16 + lane * 4;
                assert_eq!(
                    &packed[offset..offset + 4],
                    &expected,
                    "mismatch at word={word}, lane={lane}"
                );
            }
        }
    }

    #[test]
    #[should_panic(expected = "decompressed.len() must be exactly BLOCK_LEN")]
    fn compress_panics_on_wrong_length() {
        let values = vec![0u32; Scalar::BLOCK_LEN - 1];
        let mut compressed = vec![0u8; 1024];
        Scalar::compress(&values, &mut compressed, 4);
    }

    #[test]
    #[should_panic(expected = "decompressed.len() must be exactly BLOCK_LEN")]
    fn decompress_panics_on_wrong_length() {
        let compressed = vec![0u8; 1024];
        let mut decompressed = vec![0u32; Scalar::BLOCK_LEN + 1];
        Scalar::decompress(&compressed, &mut decompressed, 4);
    }

    #[test]
    #[should_panic(expected = "num_bits must be in 1..=32")]
    fn compress_panics_on_num_bits_zero() {
        let values = block(0);
        let mut compressed = vec![0u8; 1024];
        Scalar::compress(&values, &mut compressed, 0);
    }

    #[test]
    #[should_panic(expected = "num_bits must be in 1..=32")]
    fn compress_panics_on_num_bits_33() {
        let values = block(0);
        let mut compressed = vec![0u8; 1024];
        Scalar::compress(&values, &mut compressed, 33);
    }

    #[test]
    #[should_panic(expected = "compressed buffer too small")]
    fn compress_panics_on_undersized_output() {
        let values = block(u32::MAX);
        let mut compressed = vec![0u8; 1];
        Scalar::compress(&values, &mut compressed, 32);
    }

    #[test]
    #[should_panic(expected = "compressed buffer too small")]
    fn decompress_panics_on_undersized_input() {
        let compressed = vec![0u8; 1];
        let mut decompressed = vec![0u32; Scalar::BLOCK_LEN];
        Scalar::decompress(&compressed, &mut decompressed, 32);
    }

    proptest! {
        #![proptest_config(gen::proptest_config(1000))]

        #[test]
        fn prop_compress_matches_oracle(
            (num_bits, values) in gen::num_bits_and_block_values(Scalar::BLOCK_LEN)
        ) {
            let mut compressed = vec![0u8; packed_len_bytes(Scalar::BLOCK_LEN, num_bits)];
            Scalar::compress(&values, &mut compressed, num_bits);

            let expected = oracle::oracle_encode_interleaved(&values, num_bits);
            prop_assert_eq!(compressed, expected);
        }

        #[test]
        fn prop_round_trip(
            (num_bits, values) in gen::num_bits_and_block_values(Scalar::BLOCK_LEN)
        ) {
            let mut compressed = vec![0u8; packed_len_bytes(Scalar::BLOCK_LEN, num_bits)];
            Scalar::compress(&values, &mut compressed, num_bits);

            let mut decompressed = vec![0u32; Scalar::BLOCK_LEN];
            Scalar::decompress(&compressed, &mut decompressed, num_bits);
            prop_assert_eq!(decompressed, values);
        }

        #[test]
        fn prop_decompress_matches_oracle(
            (num_bits, values) in gen::num_bits_and_block_values(Scalar::BLOCK_LEN)
        ) {
            let compressed = oracle::oracle_encode_interleaved(&values, num_bits);

            let mut decompressed = vec![0u32; Scalar::BLOCK_LEN];
            Scalar::decompress(&compressed, &mut decompressed, num_bits);
            prop_assert_eq!(decompressed, values);
        }
    }
}