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
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

use crunchy::unroll;

use super::{scalar, CodingDescriptor1248};
use crate::arch::neon::{data_len8, decode_shuffle_table, encode_shuffle_table};
use crate::coding_descriptor::CodingDescriptor;
use crate::raw_group::RawGroup;
use std::arch::aarch64::{
    uint32x4_t, uint64x2_t, uint8x16_t, uint8x16x2_t, vaddl_high_u32, vaddl_u32, vaddlvq_u32,
    vaddq_u32, vaddq_u64, vaddvq_u32, vaddvq_u64, vclzq_u32, vdupq_laneq_u64, vdupq_n_u32,
    vdupq_n_u64, vextq_u32, vextq_u64, vget_low_u32, vgetq_lane_u64, vld1q_s32, vld1q_u64,
    vld1q_u8, vmovn_high_u64, vmovn_u64, vqmovn_high_u64, vqmovn_u64, vqtbl1q_u8, vqtbl2q_u8,
    vreinterpretq_s32_u8, vreinterpretq_u32_u64, vreinterpretq_u32_u8, vreinterpretq_u64_u8,
    vreinterpretq_u8_u32, vreinterpretq_u8_u64, vshlq_u32, vshrq_n_u32, vst1q_u64, vst1q_u8,
    vsubq_u64, vuzp2q_u32,
};

const ENCODE_TABLE: [[u8; 32]; 256] =
    encode_shuffle_table::<{ std::mem::size_of::<u64>() }, 32>(CodingDescriptor1248::TAG_LEN);
const DECODE_TABLE: [[u8; 32]; 256] =
    decode_shuffle_table::<{ std::mem::size_of::<u64>() }, 32>(CodingDescriptor1248::TAG_LEN);

/// Load a single 32-byte entry from table based on tag.
#[inline(always)]
unsafe fn load_shuffle(table: &[[u8; 32]; 256], tag: u8) -> (uint8x16_t, uint8x16_t) {
    let ptr = table[tag as usize].as_ptr();
    (vld1q_u8(ptr), vld1q_u8(ptr.add(16)))
}

/// Load a single 32-byte decode shuffle table entry based on tag, then narrow it to a 16-byte value.
/// Only use this if tag is set such that there are no 8-byte entries -- in this case it is valid
/// to shuffle into 4 32-bit entries in 1 register rather than 4 64-bit entries in 2 registers.
#[inline(always)]
unsafe fn load_decode_shuffle_narrow(tag: u8) -> uint8x16_t {
    let wshuf = load_shuffle(&DECODE_TABLE, tag);
    vreinterpretq_u8_u32(vmovn_high_u64(
        vmovn_u64(vreinterpretq_u64_u8(wshuf.0)),
        vreinterpretq_u64_u8(wshuf.1),
    ))
}

#[derive(Clone, Copy, Debug)]
pub(crate) struct RawGroupImpl(uint64x2_t, uint64x2_t);

impl RawGroupImpl {
    #[inline(always)]
    unsafe fn compute_tag(&self) -> (u8, usize) {
        // NEON does not provide clz on 64 bit lanes. Split each entry into hi and lo 32 bits values
        // (produce lo by saturating narrow), clz and sum to help compute the values.
        let lo = vqmovn_high_u64(vqmovn_u64(self.0), self.1);
        let hi = vuzp2q_u32(vreinterpretq_u32_u64(self.0), vreinterpretq_u32_u64(self.1));
        let clz_bytes = vshrq_n_u32(vaddq_u32(vclzq_u32(lo), vclzq_u32(hi)), 3);
        let value_tags = vqtbl1q_u8(
            vld1q_u8([3, 3, 3, 3, 2, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0].as_ptr()),
            vreinterpretq_u8_u32(clz_bytes),
        );
        let tag = vaddvq_u32(vshlq_u32(
            vreinterpretq_u32_u8(value_tags),
            vld1q_s32([0, 2, 4, 6].as_ptr()),
        )) as u8;
        let written =
            vaddvq_u32(vshlq_u32(vdupq_n_u32(1), vreinterpretq_s32_u8(value_tags))) as usize;
        (tag, written)
    }

    #[inline(always)]
    unsafe fn sum_deltas(&self, base: uint64x2_t) -> Self {
        let Self(a_b, c_d) = *self;
        let z = vdupq_n_u64(0);
        let z_a = vextq_u64(z, a_b, 1);
        let a_ab = vaddq_u64(z_a, a_b);
        let p = base;
        let pa_pab = vaddq_u64(p, a_ab);
        let b_c = vextq_u64(a_b, c_d, 1);
        let bc_cd = vaddq_u64(b_c, c_d);
        Self(pa_pab, vaddq_u64(pa_pab, bc_cd))
    }

    #[inline(always)]
    unsafe fn sum(&self) -> u64 {
        vaddvq_u64(vaddq_u64(self.0, self.1))
    }

    #[inline(always)]
    fn has_any_tag3(tag8: u64) -> bool {
        (tag8 & 0x5555555555555555 & (tag8 >> 1)) != 0
    }

    #[inline(always)]
    unsafe fn decode32(input: *const u8, tag: u8) -> (usize, uint32x4_t) {
        (
            Self::data_len(tag),
            vreinterpretq_u32_u8(vqtbl1q_u8(vld1q_u8(input), load_decode_shuffle_narrow(tag))),
        )
    }

    #[inline(always)]
    unsafe fn decode_deltas8_without_tag3(
        input: *const u8,
        tag8: u64,
        base: Self,
        output: *mut u64,
    ) -> (usize, Self) {
        let tags = tag8.to_le_bytes();
        let mut deltas = [vdupq_n_u32(0); 8];
        let mut bases = [0u64; 9];
        bases[0] = vgetq_lane_u64::<1>(base.1);
        let mut offset = 0usize;
        unroll! {
            for i in 0..8 {
                let (len, delta) = Self::decode32(input.add(offset), tags[i]);
                deltas[i] = delta;
                bases[i + 1] = bases[i].wrapping_add(vaddlvq_u32(delta));
                offset += len;
            }
        }
        let z = vdupq_n_u32(0);
        unroll! {
            for i in 0..8 {
                let p = vdupq_n_u64(bases[i]);
                let a_b_c_d = deltas[i];
                let z_a_b_c = vextq_u32(z, a_b_c_d, 3);
                let a_ab = vaddl_u32(vget_low_u32(z_a_b_c), vget_low_u32(a_b_c_d));
                let bc_cd = vaddl_high_u32(z_a_b_c, a_b_c_d);
                let pa_pab = vaddq_u64(p, a_ab);
                let group = Self(pa_pab, vaddq_u64(pa_pab, bc_cd));
                Self::store_unaligned(output.add(i * 4), group);
                if i == 7 {
                    return (offset, group);
                }
            }
        }

        unreachable!()
    }
}

impl RawGroup for RawGroupImpl {
    type Elem = u64;

    const TAG_LEN: [usize; 4] = CodingDescriptor1248::TAG_LEN;

    #[inline]
    fn set1(value: Self::Elem) -> Self {
        unsafe { RawGroupImpl(vdupq_n_u64(value), vdupq_n_u64(value)) }
    }

    #[inline]
    unsafe fn load_unaligned(ptr: *const Self::Elem) -> Self {
        // NB: there are two intrinsic calls but this should be translated into a single ldp instruction.
        RawGroupImpl(vld1q_u64(ptr), vld1q_u64(ptr.add(2)))
    }

    #[inline]
    unsafe fn store_unaligned(ptr: *mut Self::Elem, group: Self) {
        // NB: there are two intrinsic calls but this should be translated into a single stp instruction.
        vst1q_u64(ptr, group.0);
        vst1q_u64(ptr.add(2), group.1);
    }

    #[inline]
    unsafe fn encode(output: *mut u8, group: Self) -> (u8, usize) {
        let (tag, written) = group.compute_tag();
        let tbl_bytes = uint8x16x2_t(vreinterpretq_u8_u64(group.0), vreinterpretq_u8_u64(group.1));
        if written <= 16 {
            // The 4 input values will only produce 16 bytes of output or less so we only need a
            // single shuffle and store.
            let shuf = vld1q_u8(ENCODE_TABLE[tag as usize].as_ptr());
            vst1q_u8(output, vqtbl2q_u8(tbl_bytes, shuf));
        } else {
            let shuf = load_shuffle(&ENCODE_TABLE, tag);
            let out = (vqtbl2q_u8(tbl_bytes, shuf.0), vqtbl2q_u8(tbl_bytes, shuf.1));
            vst1q_u8(output, out.0);
            vst1q_u8(output.add(16), out.1);
        }
        (tag, written)
    }

    #[inline]
    unsafe fn encode_deltas(output: *mut u8, base: Self, group: Self) -> (u8, usize) {
        let b0 = vextq_u64(base.1, group.0, 1);
        let b1 = vextq_u64(group.0, group.1, 1);
        Self::encode(
            output,
            RawGroupImpl(vsubq_u64(group.0, b0), vsubq_u64(group.1, b1)),
        )
    }

    #[inline]
    unsafe fn decode(input: *const u8, tag: u8) -> (usize, Self) {
        let shuf = load_shuffle(&DECODE_TABLE, tag);
        let tbl_bytes = uint8x16x2_t(vld1q_u8(input), vld1q_u8(input.add(16)));
        let group_data = (vqtbl2q_u8(tbl_bytes, shuf.0), vqtbl2q_u8(tbl_bytes, shuf.1));
        (
            Self::data_len(tag),
            RawGroupImpl(
                vreinterpretq_u64_u8(group_data.0),
                vreinterpretq_u64_u8(group_data.1),
            ),
        )
    }

    #[inline]
    unsafe fn decode_deltas(input: *const u8, tag: u8, base: Self) -> (usize, Self) {
        let (len, group) = Self::decode(input, tag);
        (len, group.sum_deltas(vdupq_laneq_u64(base.1, 1)))
    }

    #[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, group) = Self::decode(input, tag);
        (len, group.sum())
    }

    #[inline]
    fn data_len8(tag8: u64) -> usize {
        data_len8(Self::TAG_LEN, tag8)
    }

    #[inline]
    unsafe fn decode_deltas8(
        input: *const u8,
        tag8: u64,
        base: Self,
        output: *mut Self::Elem,
    ) -> (usize, Self) {
        if !Self::has_any_tag3(tag8) {
            return Self::decode_deltas8_without_tag3(input, tag8, base, output);
        }

        let tags = tag8.to_le_bytes();
        let mut group_deltas = [Self(vdupq_n_u64(0), vdupq_n_u64(0)); 8];
        let mut bases = [0u64; 9];
        bases[0] = vgetq_lane_u64::<1>(base.1);
        let mut offset = 0usize;
        unroll! {
            for i in 0..8 {
                let (len, group) = Self::decode(input.add(offset), tags[i]);
                group_deltas[i] = group;
                bases[i + 1] = bases[i].wrapping_add(group.sum());
                offset += len;
            }
        }
        unroll! {
            for i in 0..8 {
                let group = group_deltas[i].sum_deltas(vdupq_n_u64(bases[i]));
                Self::store_unaligned(output.add(i * 4), group);
                if i == 7 {
                    return (offset, group);
                }
            }
        }

        unreachable!()
    }

    #[inline]
    unsafe fn skip_deltas8(input: *const u8, tag8: u64) -> (usize, Self::Elem) {
        if Self::has_any_tag3(tag8) {
            return crate::raw_group::default_skip_deltas8::<Self>(input, tag8);
        }

        let tags = tag8.to_le_bytes();
        let mut offset = 0usize;
        let mut sum = 0u64;
        unroll! {
            for i in 0..8 {
                let (len, deltas) = Self::decode32(input.add(offset), tags[i]);
                offset += len;
                sum = sum.wrapping_add(vaddlvq_u32(deltas));
            }
        }
        (offset, sum)
    }
}

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

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