hakmem 0.2.0

Hilbert and Morton curves with range covers, rank/select, Myers edit distance, PEXT/PDEP and byte lanes, built from bit tricks with property-tested laws. no_std, no dependencies.
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

//! Every level this machine has, in one run: the primitives with each
//! token's instructions against the bit-loop reference, on every carrier.
//! Where a build used to see one path per `RUSTFLAGS`, a Zen 5 now checks
//! `Portable` and `X86V3` together, and CI whatever its runner has.

use hakmem::isa::{self, Isa};
use hakmem::laws::reference;
use hakmem::{Wide, Word};

/// xorshift64, and the corner cases in front.
fn samples() -> impl Iterator<Item = u64> {
    let mut s = 0x9E37_79B9_7F4A_7C15_u64;
    [0, 1, u64::MAX, 0x8000_0000_0000_0000, 0x5555_5555_5555_5555]
        .into_iter()
        .chain(core::iter::from_fn(move || {
            s ^= s << 13;
            s ^= s >> 7;
            s ^= s << 17;
            Some(s)
        }))
        .take(200)
}

fn primitives<I: Isa, W: Word>(cpu: I, x: W, m: W) {
    assert_eq!(
        cpu.pext(x, m),
        reference::pext(x, m),
        "{cpu:?} pext {x:x} {m:x}"
    );
    assert_eq!(
        cpu.pdep(x, m),
        reference::pdep(x, m),
        "{cpu:?} pdep {x:x} {m:x}"
    );
    let even = W::splat_byte(0x55);
    assert_eq!(
        cpu.unzip(x),
        (reference::pext(x, even), reference::pext(x, even.shl(1))),
        "{cpu:?} unzip {x:x}"
    );
    assert_eq!(
        cpu.zip(x, m),
        reference::pdep(x, even).or(reference::pdep(m, even).shl(1)),
        "{cpu:?} zip {x:x} {m:x}"
    );
    assert_eq!(
        cpu.xor_scan(x),
        reference::prefix_xor(x),
        "{cpu:?} xor_scan {x:x}"
    );
    assert_eq!(
        cpu.xor_scan_down(x),
        reference::prefix_xor_from_top(x),
        "{cpu:?} xor_scan_down {x:x}"
    );
    for k in [
        0,
        1,
        x.count_ones().saturating_sub(1),
        x.count_ones(),
        W::BITS,
    ] {
        assert_eq!(
            cpu.select(x, k),
            reference::select(x, k),
            "{cpu:?} select {x:x} {k}"
        );
    }
}

fn every_carrier<I: Isa>(cpu: I) {
    let xs: Vec<u64> = samples().collect();
    for pair in xs.windows(3) {
        let [a, b, c] = [pair[0], pair[1], pair[2]];
        // Truncation is the point: the same bit patterns on every width.
        #[allow(clippy::cast_possible_truncation)]
        {
            primitives(cpu, a as u8, b as u8);
            primitives(cpu, a as u16, b as u16);
            primitives(cpu, a as u32, b as u32);
        }
        primitives(cpu, a, b);
        primitives(
            cpu,
            u128::from(a) << 64 | u128::from(c),
            u128::from(b) << 64 | u128::from(a),
        );
        primitives(
            cpu,
            Wide::<3>::from_limbs([a, b, c]),
            Wide::<3>::from_limbs([c, a, b]),
        );
    }
}

#[test]
fn every_level_agrees_with_the_reference() {
    let mut seen = 0;
    for level in isa::available() {
        hakmem::dispatch!(level, |cpu| every_carrier(cpu));
        seen += 1;
    }
    assert!(seen >= 1);
}

/// The plain methods are `Native`, and `Native` is a token like any other.
#[test]
fn native_is_the_plain_methods() {
    for x in samples() {
        let m = x.rotate_left(17);
        assert_eq!(isa::Native.pext(x, m), x.pext(m));
        assert_eq!(isa::Native.select(x, 3), hakmem::Bits::select(x, 3));
    }
}

/// The lanes of every level against the per-lane definitions: in a build
/// without flags this is the only place the SSSE3 carrier runs.
fn lanes_of<I: Isa>(cpu: I) {
    use hakmem::lanes::Lanes;
    use hakmem::laws;
    let table = [
        9, 8, 7, 6, 5, 4, 3, 2, 1, 0, 0x80, 0x7F, 0xFF, 0x10, 0x20, 0x30,
    ];
    let texts: [&[u8; 16]; 3] = [b"hello, world!\t\n\r", b"{\"a\": [1, 2]}   ", &[0xFF; 16]];
    for x in texts {
        for y in texts {
            let (a, b) = (I::U8x16::load(cpu, x), I::U8x16::load(cpu, y));
            for n in 0..9 {
                assert!(laws::lanes_match_reference(a, b, n, table), "{cpu:?} n={n}");
            }
            assert!(laws::lut16_composes(a, table, [0xFF; 16]), "{cpu:?}");
            assert_eq!(a.isa(), cpu);
        }
    }
}

#[test]
fn every_level_has_lawful_lanes() {
    for level in isa::available() {
        hakmem::dispatch!(level, |cpu| lanes_of(cpu));
    }
}

/// The slice methods under every level agree with `Portable`, and the
/// plain ones (which pick a level per call) with both.
fn slices_of<I: Isa>(cpu: I, xs: &[u64]) {
    use hakmem::Words;
    let p = isa::Portable;
    assert_eq!(xs.count_ones_in(cpu), xs.count_ones_in(p), "{cpu:?}");
    assert_eq!(xs.count_ones(), xs.count_ones_in(p));
    let bits = xs.len() * 64;
    for i in [
        0,
        1,
        63,
        64,
        65,
        bits / 2,
        bits.saturating_sub(1),
        bits,
        bits + 7,
    ] {
        assert_eq!(xs.rank_in(i, cpu), xs.rank_in(i, p), "{cpu:?} rank {i}");
        assert_eq!(xs.rank(i), xs.rank_in(i, p));
    }
    let n = xs.count_ones_in(p);
    for k in [0, 1, n / 3, n / 2, n.saturating_sub(1), n, n + 1] {
        assert_eq!(
            xs.select_in(k, cpu),
            xs.select_in(k, p),
            "{cpu:?} select {k}"
        );
        assert_eq!(xs.select(k), xs.select_in(k, p));
    }
    let (mut a, mut b) = (Vec::new(), Vec::new());
    xs.for_each_position_in(cpu, |i| a.push(i));
    xs.for_each_position(|i| b.push(i));
    assert_eq!(a, xs.positions().collect::<Vec<_>>(), "{cpu:?}");
    assert_eq!(b, a);
}

#[test]
fn every_level_agrees_on_slices() {
    let xs: Vec<u64> = samples().take(37).collect();
    for level in isa::available() {
        for len in [0, 1, 2, 5, 37] {
            hakmem::dispatch!(level, |cpu| slices_of(cpu, &xs[..len]));
        }
    }
}

/// `Rank9` queries under every level agree with the slice scans, on a
/// dense and a sparse bitmap, with and without the select inventory.
fn rank9_of<I: Isa>(cpu: I, bits: &[u64]) {
    use hakmem::Words;
    use hakmem::rank9::Rank9;
    let mut counts = vec![0; Rank9::counts_len(bits.len())];
    let mut select = vec![0; Rank9::select_len(bits.len())];
    let dir = Rank9::build(bits, &mut counts, &mut select);
    let mut counts2 = vec![0; Rank9::counts_len(bits.len())];
    let rank_only = Rank9::build(bits, &mut counts2, &mut []);
    let n = bits.count_ones();
    let len = bits.len() * 64;
    for i in (0..=len + 64).step_by(61) {
        assert_eq!(dir.rank_in(i, cpu), bits.rank(i), "{cpu:?} rank {i}");
        assert_eq!(dir.rank(i), bits.rank(i));
    }
    for k in (0..n + 2).step_by(7) {
        assert_eq!(dir.select_in(k, cpu), bits.select(k), "{cpu:?} select {k}");
        assert_eq!(
            rank_only.select_in(k, cpu),
            bits.select(k),
            "{cpu:?} search {k}"
        );
        assert_eq!(dir.select(k), bits.select(k));
    }
}

#[test]
fn every_level_agrees_on_rank9() {
    let dense: Vec<u64> = samples().take(150).collect();
    let sparse: Vec<u64> = samples().take(150).map(|x| 1 << (x % 64)).collect();
    for level in isa::available() {
        hakmem::dispatch!(level, |cpu| rank9_of(cpu, &dense));
        hakmem::dispatch!(level, |cpu| rank9_of(cpu, &sparse));
    }
}