streamvbyte64 0.2.0

Implementation of stream-vbyte codec supporting 64-bit wide values
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

use crunchy::unroll;

use super::{scalar, CodingDescriptor1248};
use crate::arch::shuffle::{decode_shuffle_entry, encode_shuffle_entry};
use crate::coding_descriptor::CodingDescriptor;
use crate::raw_group::RawGroup;
use std::arch::x86_64::{
    __m128i, _mm_add_epi32, _mm_add_epi64, _mm_alignr_epi8, _mm_bslli_si128, _mm_bsrli_si128,
    _mm_cvtsi128_si64x, _mm_loadu_si128, _mm_min_epu16, _mm_min_epu8, _mm_movemask_epi8,
    _mm_packus_epi16, _mm_packus_epi32, _mm_set1_epi32, _mm_set1_epi64x, _mm_shuffle_epi32,
    _mm_shuffle_epi8, _mm_storeu_si128, _mm_sub_epi64,
};

/// Build a 16 entry encode table processing half of the input group (2 entries).
/// This is necessary because pshufb can only address 16 bytes of input at a time.
const ENCODE_TABLE: [[u8; 16]; 16] = {
    let mut table = [[0u8; 16]; 16];
    let mut tag = 0usize;
    while tag < 16 {
        table[tag] = encode_shuffle_entry::<{ std::mem::size_of::<u64>() }, 16>(
            tag as u8,
            CodingDescriptor1248::TAG_LEN,
            128,
        );
        tag += 1;
    }
    table
};

/// Build a 16 entry decode table processing half of the input group (2 entries).
/// This is necessary because pshufb can only address 16 bytes of input at a time.
const DECODE_TABLE: [[u8; 16]; 16] = {
    let mut table = [[0u8; 16]; 16];
    let mut tag = 0usize;
    while tag < 16 {
        table[tag] = decode_shuffle_entry::<{ std::mem::size_of::<u64>() }, 16>(
            tag as u8,
            CodingDescriptor1248::TAG_LEN,
            128,
        );
        tag += 1;
    }
    table
};

#[inline(always)]
unsafe fn load_shuffle(table: &[[u8; 16]; 16], nibble_tags: (usize, usize)) -> (__m128i, __m128i) {
    (
        _mm_loadu_si128(table[nibble_tags.0].as_ptr() as *const __m128i),
        _mm_loadu_si128(table[nibble_tags.1].as_ptr() as *const __m128i),
    )
}

// TODO: this could be shared with generate_nibble_tag_len_table().
const NIBBLE_LEN: [usize; 16] = [2, 3, 5, 9, 3, 4, 6, 10, 5, 6, 8, 12, 9, 10, 12, 16];

#[derive(Clone, Copy, Debug)]
pub(crate) struct RawGroupImpl(__m128i, __m128i);

impl RawGroupImpl {
    /// Use a similar approach to 1234 ssse3 impl in that we movemask to generate the tag value.
    /// Do 16-bit saturating narrow, a bit of manipulation, followed by 32-bit saturating narrow.
    /// The bit manipulation ensures that we handle cases with bytes 2-4 set correctly, the 32-bit
    /// narrow will ensure that if bytes 5-8 are set we will set all bits in the output tag.
    #[inline(always)]
    unsafe fn compute_tag(&self) -> usize {
        let mmask = _mm_set1_epi64x(0x0101010101010100);
        let m = (_mm_min_epu8(self.0, mmask), _mm_min_epu8(self.1, mmask));
        let n1 = _mm_packus_epi16(m.0, m.1);
        let m1 = _mm_min_epu16(n1, _mm_set1_epi32(0x1_0100));
        // map 0x0100 => 0x8000 for movemask
        let m2 = _mm_add_epi32(m1, _mm_set1_epi32(0x7f00));
        let n2 = _mm_packus_epi32(m2, m2);
        _mm_movemask_epi8(n2) as usize & 0xff
    }

    /// Splits input 8-bit `tag` into two nibble-length tags covering two entries instead of 4.
    #[inline(always)]
    fn nibble_tags(tag: usize) -> (usize, usize) {
        debug_assert!(tag < 256);
        (tag & 0xf, tag >> 4)
    }

    /// Return the data length for each nibble tag.
    #[inline(always)]
    fn nibble_data_len(nibble_tags: (usize, usize)) -> (usize, usize) {
        (NIBBLE_LEN[nibble_tags.0], NIBBLE_LEN[nibble_tags.1])
    }
}

impl RawGroup for RawGroupImpl {
    type Elem = u64;
    const TAG_LEN: [usize; 4] = CodingDescriptor1248::TAG_LEN;

    #[inline]
    fn set1(value: Self::Elem) -> Self {
        unsafe {
            let h = _mm_set1_epi64x(std::mem::transmute(value));
            RawGroupImpl(h, h)
        }
    }

    #[inline]
    unsafe fn load_unaligned(ptr: *const Self::Elem) -> Self {
        RawGroupImpl(
            _mm_loadu_si128(ptr as *const __m128i),
            _mm_loadu_si128(ptr.add(2) as *const __m128i),
        )
    }

    #[inline]
    unsafe fn store_unaligned(ptr: *mut Self::Elem, group: Self) {
        _mm_storeu_si128(ptr as *mut __m128i, group.0);
        _mm_storeu_si128(ptr.add(2) as *mut __m128i, group.1);
    }

    #[inline]
    unsafe fn encode(output: *mut u8, group: Self) -> (u8, usize) {
        let tag = group.compute_tag();
        let nibble_tags = Self::nibble_tags(tag);
        let nibble_data_len = Self::nibble_data_len(nibble_tags);
        let shuf = load_shuffle(&ENCODE_TABLE, nibble_tags);
        _mm_storeu_si128(output as *mut __m128i, _mm_shuffle_epi8(group.0, shuf.0));
        _mm_storeu_si128(
            output.add(nibble_data_len.0) as *mut __m128i,
            _mm_shuffle_epi8(group.1, shuf.1),
        );
        (tag as u8, nibble_data_len.0 + nibble_data_len.1)
    }

    #[inline]
    unsafe fn encode_deltas(output: *mut u8, base: Self, group: Self) -> (u8, usize) {
        let delta_base = (
            _mm_alignr_epi8::<8>(group.0, base.1),
            _mm_alignr_epi8::<8>(group.1, group.0),
        );
        let delta_group = RawGroupImpl(
            _mm_sub_epi64(group.0, delta_base.0),
            _mm_sub_epi64(group.1, delta_base.1),
        );
        Self::encode(output, delta_group)
    }

    #[inline]
    unsafe fn decode(input: *const u8, tag: u8) -> (usize, Self) {
        let nibble_tags = Self::nibble_tags(tag as usize);
        let nibble_data_len = Self::nibble_data_len(nibble_tags);
        let inputs = (
            _mm_loadu_si128(input as *const __m128i),
            _mm_loadu_si128(input.add(nibble_data_len.0) as *const __m128i),
        );
        let shuf = load_shuffle(&DECODE_TABLE, nibble_tags);
        (
            nibble_data_len.0 + nibble_data_len.1,
            RawGroupImpl(
                _mm_shuffle_epi8(inputs.0, shuf.0),
                _mm_shuffle_epi8(inputs.1, shuf.1),
            ),
        )
    }

    #[inline]
    unsafe fn decode_deltas(input: *const u8, tag: u8, base: Self) -> (usize, Self) {
        let delta_base = _mm_shuffle_epi32(base.1, 0b11101110);
        let (len, Self(a_b, c_d)) = Self::decode(input, tag);
        let z_a = _mm_bslli_si128(a_b, 8);
        let b_c = _mm_alignr_epi8(c_d, a_b, 8);
        let a_ab = _mm_add_epi64(z_a, a_b);
        let pa_pab = _mm_add_epi64(delta_base, a_ab);
        let bc_cd = _mm_add_epi64(b_c, c_d);
        (len, RawGroupImpl(pa_pab, _mm_add_epi64(pa_pab, bc_cd)))
    }

    #[inline]
    fn data_len(tag: u8) -> usize {
        scalar::RawGroupImpl::data_len(tag)
    }

    #[inline]
    unsafe fn skip_deltas(input: *const u8, tag: u8) -> (usize, Self::Elem) {
        let (len, Self(a_b, c_d)) = Self::decode(input, tag);
        let ac_bd = _mm_add_epi64(a_b, c_d);
        let abcd_bd = _mm_add_epi64(_mm_bsrli_si128::<8>(ac_bd), ac_bd);
        (len, std::mem::transmute(_mm_cvtsi128_si64x(abcd_bd)))
    }

    #[inline]
    unsafe fn skip_deltas8(input: *const u8, tag8: u64) -> (usize, Self::Elem) {
        let tags = tag8.to_le_bytes();
        let (mut offset, group) = Self::decode(input, tags[0]);
        let mut sum_delta = _mm_add_epi64(group.0, group.1);
        unroll! {
            for i in 1..8 {
                let (len, group) = Self::decode(input.add(offset), tags[i]);
                offset += len;
                sum_delta = _mm_add_epi64(sum_delta, group.0);
                sum_delta = _mm_add_epi64(sum_delta, group.1);
            }
        }
        sum_delta = _mm_add_epi64(sum_delta, _mm_bsrli_si128(sum_delta, 8));
        (offset, std::mem::transmute(_mm_cvtsi128_si64x(sum_delta)))
    }
}

#[cfg(test)]
crate::tests::raw_group_test_suite!();

#[cfg(test)]
crate::tests::compat_test_suite!();