fashex 0.0.13

Hexadecimal string encoding and decoding with best-effort SIMD acceleration.
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

//! Optimized implementations based on unstable [`core::simd`] APIs.

#![allow(clippy::similar_names, reason = "XXX")]

use core::mem::MaybeUninit;
use core::simd::prelude::*;
use core::{ptr, slice};

use crate::backend::generic::{decode_generic_unchecked, encode_generic_unchecked};
use crate::error::InvalidInput;
use crate::util::lut16;

#[cfg_attr(
    any(
        target_arch = "x86",
        target_arch = "x86_64",
        target_arch = "aarch64",
        target_arch = "arm64ec",
        target_arch = "loongarch32",
        target_arch = "loongarch64",
        target_arch = "wasm32"
    ),
    allow(dead_code, reason = "XXX")
)]
/// ## Safety
///
/// We assume that:
///
/// 1. `src.len() <= dst.len()`.
pub(crate) unsafe fn encode_simd128_unchecked<const UPPER: bool>(
    src: &[u8],
    dst: &mut [[MaybeUninit<u8>; 2]],
) {
    #[allow(clippy::identity_op, reason = "XXX")]
    /// Process 16 bytes of input, and produce 16 * 2 bytes of output.
    const BATCH_ELEMS_V128_X1: usize = size_of::<u8x16>() * 1;

    debug_assert!(src.len() <= dst.len());

    if src.len() >= BATCH_ELEMS_V128_X1 {
        let mask = u8x16::splat(0b_0000_1111);

        let lut = u8x16::from_array(*lut16::<UPPER>());

        let encode_v128 = |src: *const u8x16, dst: *mut u8x16| {
            let chunk = src.read_unaligned();

            // let [hi; 16] = [(byte >> 4) & 0b_0000_1111; 16];
            let hi = chunk >> 4;
            // let [lo; 16] = [byte & 0b_0000_1111; 16];
            let lo = chunk & mask;

            // interleave(hi, lo)
            let (a, b) = u8x16::interleave(hi, lo);

            // let [char; 16] = [lut[a]; 16];
            let out0 = lut.swizzle_dyn(a);
            // let [char; 16] = [lut[b]; 16];
            let out1 = lut.swizzle_dyn(b);

            ptr::copy_nonoverlapping(out0.as_array().as_ptr(), dst.add(0).cast(), out0.len());
            ptr::copy_nonoverlapping(out1.as_array().as_ptr(), dst.add(1).cast(), out1.len());
        };

        let batches = src.len() / BATCH_ELEMS_V128_X1;
        let remainder = src.len() % BATCH_ELEMS_V128_X1;

        for i in 0..batches {
            encode_v128(
                src.as_ptr().cast::<u8x16>().add(i),
                dst.as_mut_ptr().cast::<u8x16>().add(i * 2),
            );
        }

        encode_generic_unchecked::<UPPER>(
            slice::from_raw_parts(src.as_ptr().add(batches * BATCH_ELEMS_V128_X1), remainder),
            slice::from_raw_parts_mut(
                dst.as_mut_ptr().add(batches * BATCH_ELEMS_V128_X1),
                remainder,
            ),
        );
    } else {
        encode_generic_unchecked::<UPPER>(src, dst);
    }
}

#[cfg_attr(
    any(
        target_arch = "x86",
        target_arch = "x86_64",
        target_arch = "aarch64",
        target_arch = "arm64ec",
        target_arch = "loongarch32",
        target_arch = "loongarch64",
        target_arch = "wasm32"
    ),
    allow(dead_code, reason = "XXX")
)]
#[allow(clippy::cast_possible_wrap, reason = "XXX")]
/// ## Safety
///
/// We assume that:
///
/// 1. `src.len() <= dst.len()`.
pub(crate) unsafe fn decode_simd128_unchecked(
    src: *const [[u8; 2]],
    dst: *mut [MaybeUninit<u8>],
) -> Result<(), InvalidInput> {
    /// Process 2 * 8 * 2 bytes of input, and produce 2 * 8 bytes of output.
    const BATCH_ELEMS_V128_X2: usize = size_of::<i8x16>() / 2 * 2;

    if src.len() >= BATCH_ELEMS_V128_X2 {
        let n_c6 = i8x16::splat((0xFF_u8 - b'9').cast_signed());
        let n_06 = i8x16::splat(0x06);
        let n_f0 = i8x16::splat(0xF0_u8.cast_signed());

        let n_df = i8x16::splat(0xDF_u8.cast_signed());
        let u_a = i8x16::splat(b'A'.cast_signed());
        let n_0a = i8x16::splat(0x0A);

        let n_0f = u8x16::splat(15);

        let batches = src.len() / BATCH_ELEMS_V128_X2;
        let remainder = src.len() % BATCH_ELEMS_V128_X2;

        let mut invalid = false;

        let decode_chunk = |chunk: u8x16| -> u8x16 {
            // Digits '0'..'9' → 0..9, others > 15.
            let d = (chunk.cast::<i8>() + n_c6)
                .cast::<u8>()
                .saturating_sub(n_06.cast::<u8>())
                .cast::<i8>()
                - n_f0;
            // Letters 'A'..'F'/'a'..'f' → 10..15, others > 15.
            let a = ((chunk.cast::<i8>() & n_df) - u_a)
                .cast::<u8>()
                .saturating_add(n_0a.cast::<u8>());
            // Valid nibble wins (0..15), invalid stays > 15.
            d.cast::<u8>().simd_min(a)
        };

        let mut decode_v128x2 = |src: *const u8x16, dst: *mut u8x16| {
            let chunk0 = src.add(0).read_unaligned();
            let chunk1 = src.add(1).read_unaligned();

            let n0 = decode_chunk(chunk0);
            let n1 = decode_chunk(chunk1);

            // Validate: saturating add sets MSB if nibble > 15.
            invalid |= (n0 | n1).simd_gt(n_0f).any();

            let out = {
                let (hi, lo) = Simd::deinterleave(n0, n1);
                (hi << 4) | lo
            };

            ptr::copy_nonoverlapping(out.as_array().as_ptr(), dst.cast::<u8>(), out.len());
        };

        for i in 0..batches {
            decode_v128x2(src.cast::<u8x16>().add(i * 2), dst.cast::<u8x16>().add(i));
        }

        if invalid {
            return Err(InvalidInput);
        }

        decode_generic_unchecked::<false>(
            ptr::slice_from_raw_parts(
                src.cast::<[u8; 2]>().add(batches * BATCH_ELEMS_V128_X2),
                remainder,
            ),
            ptr::slice_from_raw_parts_mut(
                dst.cast::<MaybeUninit<u8>>()
                    .add(batches * BATCH_ELEMS_V128_X2),
                remainder,
            ),
        )
    } else {
        decode_generic_unchecked::<false>(src, dst)
    }
}

#[cfg(test)]
mod smoking {
    use super::*;
    use crate::backend::tests::{
        check_decode_validation_any_backend, check_encode_decode_any_backend,
    };

    #[test]
    fn test_encode_decode_simd128() {
        check_encode_decode_any_backend::<true>(
            encode_simd128_unchecked::<true>,
            decode_generic_unchecked::<false>,
        );
        check_encode_decode_any_backend::<false>(
            encode_simd128_unchecked::<false>,
            decode_generic_unchecked::<false>,
        );
        check_encode_decode_any_backend::<true>(
            encode_simd128_unchecked::<true>,
            decode_simd128_unchecked,
        );
        check_encode_decode_any_backend::<false>(
            encode_simd128_unchecked::<false>,
            decode_simd128_unchecked,
        );
    }

    #[test]
    fn test_decode_validation_simd128() {
        check_decode_validation_any_backend(decode_simd128_unchecked);
    }
}