csv_lib 1.0.7

Library for parsing CSV files using memory-mapped I/O, with low alloc, and AVX2/NEON support
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

use memchr::{memchr, memchr2, memchr3, memrchr};
//=================================================================//
//=====================   PUBLIC  FUNCTIONS   =====================//
//=================================================================//

/// ## Locate Line Break using Memchr3
/// - Locates the line break, using memchr2. Is compatible with CPU made before 2013. (Don't have AVX2 compatibility)
pub(crate) fn locate_line_break_memchr3(slice: &[u8], cursor: usize, separator: u8) -> usize {
    let r = if separator == b'\r' || separator == b'\n' {
        memchr2(b'\n', b'\r', slice).map(|i| {
            if slice[i] == b'\r' && slice.get(i + 1) == Some(&b'\n') {
                cursor + i + 2
            } else {
                cursor + i + 1
            }
        })
    } else {
        memchr3(b'\n', b'\r', separator, slice).map(|i| {
            if slice[i] == b'\r' && slice.get(i + 1) == Some(&b'\n') {
                cursor + i + 2
            } else {
                cursor + i + 1
            }
        })
    };
    r.unwrap_or_else(|| 0)
}

/// ## Find Delimiter Bounds
/// - Finds the first and last occurrence of a delimiter in a slice.
/// - Uses SIMD under the hood via memchr and memrchr.
pub(crate) fn find_delimiter_bounds(slice: &[u8], delimiter: u8) -> Option<(usize, usize)> {
    let first = memchr(delimiter, slice)?;
    let last = memrchr(delimiter, slice)?;
    Some((first, last))
}

/// ## Get Cleaned Slice
/// - Returns a slice without the delimiters if they are at the first and last position.
pub(crate) fn get_cleaned(slice: &[u8], delimiter: u8) -> &[u8] {
    if delimiter == 0 || slice.len() < 2 {
        return slice;
    }
    if let Some((first, last)) = find_delimiter_bounds(slice, delimiter) {
        return &slice[first + 1..last];
    }
    slice
}

/// ## Locate Line Break AVX2
/// - Finds the next line break, using AVX2 feature.
#[cfg(target_arch = "x86_64")]
#[target_feature(enable = "avx2")]
pub(crate) unsafe fn locate_line_break_avx2(buffer: &[u8], separator: u8) -> usize {
    unsafe {
        use std::arch::x86_64::*;

        let mut i = 0;
        let len = buffer.len();

        let pattern_n = _mm256_set1_epi8(b'\n' as i8);
        let pattern_r = _mm256_set1_epi8(b'\r' as i8);
        let check_separator = separator != b'\r' && separator != b'\n';
        let pattern_sep = if check_separator {
            Some(_mm256_set1_epi8(separator as i8))
        } else {
            None
        };

        while i + 32 <= len {
            let chunk = _mm256_loadu_si256(buffer.as_ptr().add(i) as *const __m256i);

            let cmp_n = _mm256_cmpeq_epi8(chunk, pattern_n);
            let cmp_r = _mm256_cmpeq_epi8(chunk, pattern_r);
            let cmp_sep = if let Some(pattern_sep) = pattern_sep {
                Some(_mm256_cmpeq_epi8(chunk, pattern_sep))
            } else {
                None
            };

            let mask_n = _mm256_movemask_epi8(cmp_n);
            let mask_r = _mm256_movemask_epi8(cmp_r);
            let mask_sep = cmp_sep.map_or(0, |c| _mm256_movemask_epi8(c));

            if mask_n != 0 || mask_r != 0 || mask_sep != 0 {
                for j in 0..32 {
                    let pos = i + j;
                    if pos >= len {
                        break;
                    }
                    match buffer[pos] {
                        b'\r' => {
                            if pos + 1 < len && buffer[pos + 1] == b'\n' {
                                return pos + 2;
                            } else {
                                return pos + 1;
                            }
                        }
                        b'\n' => return pos + 1,
                        byte if check_separator && byte == separator => return pos + 1,
                        _ => {}
                    }
                }
            }

            i += 32;
        }

        while i < len {
            match buffer[i] {
                b'\r' if i + 1 < len && buffer[i + 1] == b'\n' => return i + 2,
                b'\r' | b'\n' => return i + 1,
                byte if check_separator && byte == separator => return i + 1,
                _ => i += 1,
            }
        }

        buffer.len()
    }
}

/// ## Locate Line Break NEON
///
/// - Finds the next line break, using aarch64 feature NEON.
#[cfg(target_arch = "aarch64")]
#[allow(dead_code, unsafe_code)]
pub(crate) unsafe fn locate_line_break_neon(buffer: &[u8], separator: u8) -> usize {
    unsafe {
        use std::arch::aarch64::*;

        let mut i = 0;
        let len = buffer.len();

        let pattern_n = vdupq_n_u8(b'\n');
        let pattern_r = vdupq_n_u8(b'\r');
        // Only create separator pattern if separator is different from \r and \n
        let check_separator = separator != b'\r' && separator != b'\n';
        let pattern_sep = if check_separator {
            Some(vdupq_n_u8(separator))
        } else {
            None
        };

        while i + 16 <= len {
            let chunk = vld1q_u8(buffer.as_ptr().add(i));
            let cmp_n = vceqq_u8(chunk, pattern_n);
            let cmp_r = vceqq_u8(chunk, pattern_r);
            let cmp_sep = if let Some(pattern_sep) = pattern_sep {
                Some(vceqq_u8(chunk, pattern_sep))
            } else {
                None
            };

            let mask_n = vmaxvq_u8(cmp_n);
            let mask_r = vmaxvq_u8(cmp_r);
            let mask_sep = cmp_sep.map_or(0, |c| vmaxvq_u8(c));

            if mask_n != 0 || mask_r != 0 || mask_sep != 0 {
                let mut tmp_n = [0u8; 16];
                let mut tmp_r = [0u8; 16];
                let mut tmp_sep = [0u8; 16];

                vst1q_u8(tmp_n.as_mut_ptr(), cmp_n);
                vst1q_u8(tmp_r.as_mut_ptr(), cmp_r);
                if let Some(cmp) = cmp_sep {
                    vst1q_u8(tmp_sep.as_mut_ptr(), cmp);
                }

                for j in 0..16 {
                    if tmp_r[j] == 0xFF {
                        if i + j + 1 < len && buffer[i + j + 1] == b'\n' {
                            return i + j + 2; // \r\n combo
                        } else {
                            return i + j + 1; // Solo \r
                        }
                    } else if tmp_n[j] == 0xFF {
                        return i + j + 1; // Solo \n
                    } else if check_separator && tmp_sep[j] == 0xFF {
                        return i + j + 1; // Custom separator
                    }
                }
            }

            i += 16;
        }
        while i < len {
            match buffer[i] {
                b'\r' if i + 1 < len && buffer[i + 1] == b'\n' => return i + 2,
                b'\r' | b'\n' => return i + 1,
                byte if check_separator && byte == separator => return i + 1,
                _ => i += 1,
            }
        }

        buffer.len()
    }
}