packed_seq/
ascii.rs

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
use super::*;

/// Maps ASCII to `[0, 4)` on the fly.
/// Prefer first packing into a `PackedSeqVec` for storage.
impl<'s> Seq<'s> for &[u8] {
    const BASES_PER_BYTE: usize = 1;
    const BITS_PER_CHAR: usize = 8;
    type SeqVec = Vec<u8>;

    #[inline(always)]
    fn len(&self) -> usize {
        <[u8]>::len(self)
    }

    #[inline(always)]
    fn get(&self, index: usize) -> u8 {
        self[index]
    }

    #[inline(always)]
    fn get_ascii(&self, index: usize) -> u8 {
        self[index]
    }

    #[inline(always)]
    fn to_word(&self) -> usize {
        assert!(self.len() <= usize::BITS as usize / 8);
        let mask = usize::MAX >> (64 - 8 * self.len());
        unsafe { (self.as_ptr() as *const usize).read_unaligned() & mask }
    }

    /// Convert to an owned version.
    fn to_vec(&self) -> Vec<u8> {
        <[u8]>::to_vec(self)
    }

    #[inline(always)]
    fn slice(&self, range: Range<usize>) -> Self {
        &self[range]
    }

    /// Iter the ASCII characters.
    #[inline(always)]
    fn iter_bp(self) -> impl ExactSizeIterator<Item = u8> + Clone {
        self.iter().copied()
    }

    /// Iter the ASCII characters in parallel.
    #[inline(always)]
    fn par_iter_bp(self, context: usize) -> (impl ExactSizeIterator<Item = S> + Clone, Self) {
        let num_kmers = self.len().saturating_sub(context - 1);
        let n = num_kmers / L;

        let base_ptr = self.as_ptr();
        let offsets_lanes_0_4: u64x4 = from_fn(|l| (l * n) as u64).into();
        let offsets_lanes_4_8: u64x4 = from_fn(|l| ((4 + l) * n) as u64).into();
        let mut upcoming_1 = S::ZERO;
        let mut upcoming_2 = S::ZERO;

        let par_len = if num_kmers == 0 { 0 } else { n + context - 1 };
        let it = (0..par_len).map(move |i| {
            if i % 4 == 0 {
                if i % 8 == 0 {
                    // Read a u64 containing the next 8 characters.
                    let idx_0_4 = offsets_lanes_0_4 + u64x4::splat(i as u64);
                    let idx_4_8 = offsets_lanes_4_8 + u64x4::splat(i as u64);
                    let u64_0_4: S = unsafe { transmute(intrinsics::gather(base_ptr, idx_0_4)) };
                    let u64_4_8: S = unsafe { transmute(intrinsics::gather(base_ptr, idx_4_8)) };
                    // Split into two vecs containing a u32 of 4 characters each.
                    (upcoming_1, upcoming_2) = intrinsics::deinterleave(u64_0_4, u64_4_8);
                } else {
                    // Move on to the next u32 containing 4 buffered characters.
                    upcoming_1 = upcoming_2;
                }
            }
            // Extract the last 1-byte character.
            let chars = upcoming_1 & S::splat(0xff);
            // Shift remaining characters to the right.
            upcoming_1 = upcoming_1 >> S::splat(8);
            chars
        });

        (it, &self[L * n..])
    }

    #[inline(always)]
    fn par_iter_bp_delayed(
        self,
        context: usize,
        delay: usize,
    ) -> (impl ExactSizeIterator<Item = (S, S)> + Clone, Self) {
        let num_kmers = self.len().saturating_sub(context - 1);
        let n = num_kmers / L;

        let base_ptr = self.as_ptr();
        let offsets_lanes_0_4: u64x4 = from_fn(|l| (l * n) as u64).into();
        let offsets_lanes_4_8: u64x4 = from_fn(|l| ((4 + l) * n) as u64).into();

        let mut upcoming = S::ZERO;
        let mut upcoming_d = S::ZERO;

        // Even buf_len is nice to only have the write==buf_len check once.
        // We also make it the next power of 2, for faster modulo operations.
        // delay/4: number of bp in a u32.
        let buf_len = (delay / 4 + 2).next_power_of_two();
        let buf_mask = buf_len - 1;
        let mut buf = vec![S::ZERO; buf_len];
        let mut write_idx = 0;
        // We compensate for the first delay/16 triggers of the check below that
        // happen before the delay is actually reached.
        let mut read_idx = (buf_len - delay / 4) % buf_len;

        let par_len = if num_kmers == 0 { 0 } else { n + context - 1 };
        let it = (0..par_len).map(move |i| {
            if i % 4 == 0 {
                if i % 8 == 0 {
                    // Read a u64 containing the next 8 characters.
                    let idx_0_4 = offsets_lanes_0_4 + u64x4::splat(i as u64);
                    let idx_4_8 = offsets_lanes_4_8 + u64x4::splat(i as u64);
                    let u64_0_4: S = unsafe { transmute(intrinsics::gather(base_ptr, idx_0_4)) };
                    let u64_4_8: S = unsafe { transmute(intrinsics::gather(base_ptr, idx_4_8)) };
                    // Split into two vecs containing a u32 of 4 characters each.
                    let (a, b) = intrinsics::deinterleave(u64_0_4, u64_4_8);
                    unsafe { assert_unchecked(write_idx < buf.len()) };
                    buf[write_idx] = a;
                    upcoming = a;
                    write_idx += 1;
                    unsafe { assert_unchecked(write_idx < buf.len()) };
                    buf[write_idx] = b;
                    // write_idx will be incremented one more in the `else` below.
                } else {
                    // Move on to the next u32 containing 4 buffered characters.
                    unsafe { assert_unchecked(write_idx < buf.len()) };
                    upcoming = buf[write_idx];
                    write_idx += 1;
                    write_idx &= buf_mask;
                }
            }
            if i % 4 == delay % 4 {
                unsafe { assert_unchecked(read_idx < buf.len()) };
                upcoming_d = buf[read_idx];
                read_idx += 1;
                read_idx &= buf_mask;
            }
            // Extract the last 2 bits of each character.
            let chars = upcoming & S::splat(0xff);
            let chars_d = upcoming_d & S::splat(0xff);
            // Shift remaining characters to the right.
            upcoming = upcoming >> S::splat(8);
            upcoming_d = upcoming_d >> S::splat(8);
            (chars, chars_d)
        });

        (it, &self[L * n..])
    }

    #[inline(always)]
    fn par_iter_bp_delayed_2(
        self,
        context: usize,
        delay1: usize,
        delay2: usize,
    ) -> (impl ExactSizeIterator<Item = (S, S, S)> + Clone, Self) {
        assert!(delay1 <= delay2, "Delay1 must be at most delay2.");

        let num_kmers = self.len().saturating_sub(context - 1);
        let n = num_kmers / L;

        let base_ptr = self.as_ptr();
        let offsets_lanes_0_4: u64x4 = from_fn(|l| (l * n) as u64).into();
        let offsets_lanes_4_8: u64x4 = from_fn(|l| ((4 + l) * n) as u64).into();

        let mut upcoming = S::ZERO;
        let mut upcoming_d1 = S::ZERO;
        let mut upcoming_d2 = S::ZERO;

        // Even buf_len is nice to only have the write==buf_len check once.
        // We also make it the next power of 2, for faster modulo operations.
        // delay/4: number of bp in a u32.
        let buf_len = (delay2 / 4 + 2).next_power_of_two();
        let buf_mask = buf_len - 1;
        let mut buf = vec![S::ZERO; buf_len];
        let mut write_idx = 0;
        // We compensate for the first delay/16 triggers of the check below that
        // happen before the delay is actually reached.
        let mut read_idx1 = (buf_len - delay1 / 4) % buf_len;
        let mut read_idx2 = (buf_len - delay2 / 4) % buf_len;

        let par_len = if num_kmers == 0 { 0 } else { n + context - 1 };
        let it = (0..par_len).map(move |i| {
            if i % 4 == 0 {
                if i % 8 == 0 {
                    // Read a u64 containing the next 8 characters.
                    let idx_0_4 = offsets_lanes_0_4 + u64x4::splat(i as u64);
                    let idx_4_8 = offsets_lanes_4_8 + u64x4::splat(i as u64);
                    let u64_0_4: S = unsafe { transmute(intrinsics::gather(base_ptr, idx_0_4)) };
                    let u64_4_8: S = unsafe { transmute(intrinsics::gather(base_ptr, idx_4_8)) };
                    // Split into two vecs containing a u32 of 4 characters each.
                    let (a, b) = intrinsics::deinterleave(u64_0_4, u64_4_8);
                    unsafe { assert_unchecked(write_idx < buf.len()) };
                    buf[write_idx] = a;
                    upcoming = a;
                    write_idx += 1;
                    unsafe { assert_unchecked(write_idx < buf.len()) };
                    buf[write_idx] = b;
                    // write_idx will be incremented one more in the `else` below.
                } else {
                    // Move on to the next u32 containing 4 buffered characters.
                    unsafe { assert_unchecked(write_idx < buf.len()) };
                    upcoming = buf[write_idx];
                    write_idx += 1;
                    write_idx &= buf_mask;
                }
            }
            if i % 4 == delay1 % 4 {
                unsafe { assert_unchecked(read_idx1 < buf.len()) };
                upcoming_d1 = buf[read_idx1];
                read_idx1 += 1;
                read_idx1 &= buf_mask;
            }
            if i % 4 == delay2 % 4 {
                unsafe { assert_unchecked(read_idx2 < buf.len()) };
                upcoming_d2 = buf[read_idx2];
                read_idx2 += 1;
                read_idx2 &= buf_mask;
            }
            // Extract the last 2 bits of each character.
            let chars = upcoming & S::splat(0xff);
            let chars_d1 = upcoming_d1 & S::splat(0xff);
            let chars_d2 = upcoming_d2 & S::splat(0xff);
            // Shift remaining characters to the right.
            upcoming = upcoming >> S::splat(8);
            upcoming_d1 = upcoming_d1 >> S::splat(8);
            upcoming_d2 = upcoming_d2 >> S::splat(8);
            (chars, chars_d1, chars_d2)
        });

        (it, &self[L * n..])
    }

    // TODO: This is not very optimized.
    fn cmp_lcp(&self, other: &Self) -> (std::cmp::Ordering, usize) {
        for i in 0..self.len().min(other.len()) {
            if self[i] != other[i] {
                return (self[i].cmp(&other[i]), i);
            }
        }
        (self.len().cmp(&other.len()), self.len().min(other.len()))
    }
}

impl SeqVec for Vec<u8> {
    type Seq<'s> = &'s [u8];

    /// Get the underlying ASCII text.
    fn into_raw(self) -> Vec<u8> {
        self
    }

    fn as_slice(&self) -> Self::Seq<'_> {
        self.as_slice()
    }

    fn push_seq(&mut self, seq: &[u8]) -> Range<usize> {
        let start = self.len();
        let end = start + seq.len();
        let range = start..end;
        self.extend(seq);
        range
    }

    fn push_ascii(&mut self, seq: &[u8]) -> Range<usize> {
        self.push_seq(seq)
    }

    fn random(n: usize) -> Self {
        let mut rng = rand::rng();
        (0..n).map(|_| rng.random::<u8>()).collect()
    }
}