core/slice/
memchr.rs

1// Original implementation taken from rust-memchr.
2// Copyright 2015 Andrew Gallant, bluss and Nicolas Koch
3
4use crate::intrinsics::const_eval_select;
5
6const LO_USIZE: usize = usize::repeat_u8(0x01);
7const HI_USIZE: usize = usize::repeat_u8(0x80);
8const USIZE_BYTES: usize = size_of::<usize>();
9
10/// Returns `true` if `x` contains any zero byte.
11///
12/// From *Matters Computational*, J. Arndt:
13///
14/// "The idea is to subtract one from each of the bytes and then look for
15/// bytes where the borrow propagated all the way to the most significant
16/// bit."
17#[inline]
18const fn contains_zero_byte(x: usize) -> bool {
19    x.wrapping_sub(LO_USIZE) & !x & HI_USIZE != 0
20}
21
22/// Returns the first index matching the byte `x` in `text`.
23#[inline]
24#[must_use]
25pub const fn memchr(x: u8, text: &[u8]) -> Option<usize> {
26    // Fast path for small slices.
27    if text.len() < 2 * USIZE_BYTES {
28        return memchr_naive(x, text);
29    }
30
31    memchr_aligned(x, text)
32}
33
34#[inline]
35const fn memchr_naive(x: u8, text: &[u8]) -> Option<usize> {
36    let mut i = 0;
37
38    // FIXME(const-hack): Replace with `text.iter().pos(|c| *c == x)`.
39    while i < text.len() {
40        if text[i] == x {
41            return Some(i);
42        }
43
44        i += 1;
45    }
46
47    None
48}
49
50#[rustc_allow_const_fn_unstable(const_eval_select)] // fallback impl has same behavior
51const fn memchr_aligned(x: u8, text: &[u8]) -> Option<usize> {
52    // The runtime version behaves the same as the compiletime version, it's
53    // just more optimized.
54    const_eval_select!(
55        @capture { x: u8, text: &[u8] } -> Option<usize>:
56        if const {
57            memchr_naive(x, text)
58        } else {
59            // Scan for a single byte value by reading two `usize` words at a time.
60            //
61            // Split `text` in three parts
62            // - unaligned initial part, before the first word aligned address in text
63            // - body, scan by 2 words at a time
64            // - the last remaining part, < 2 word size
65
66            // search up to an aligned boundary
67            let len = text.len();
68            let ptr = text.as_ptr();
69            let mut offset = ptr.align_offset(USIZE_BYTES);
70
71            if offset > 0 {
72                offset = offset.min(len);
73                let slice = &text[..offset];
74                if let Some(index) = memchr_naive(x, slice) {
75                    return Some(index);
76                }
77            }
78
79            // search the body of the text
80            let repeated_x = usize::repeat_u8(x);
81            while offset <= len - 2 * USIZE_BYTES {
82                // SAFETY: the while's predicate guarantees a distance of at least 2 * usize_bytes
83                // between the offset and the end of the slice.
84                unsafe {
85                    let u = *(ptr.add(offset) as *const usize);
86                    let v = *(ptr.add(offset + USIZE_BYTES) as *const usize);
87
88                    // break if there is a matching byte
89                    let zu = contains_zero_byte(u ^ repeated_x);
90                    let zv = contains_zero_byte(v ^ repeated_x);
91                    if zu || zv {
92                        break;
93                    }
94                }
95                offset += USIZE_BYTES * 2;
96            }
97
98            // Find the byte after the point the body loop stopped.
99            // FIXME(const-hack): Use `?` instead.
100            // FIXME(const-hack, fee1-dead): use range slicing
101            let slice =
102            // SAFETY: offset is within bounds
103                unsafe { super::from_raw_parts(text.as_ptr().add(offset), text.len() - offset) };
104            if let Some(i) = memchr_naive(x, slice) { Some(offset + i) } else { None }
105        }
106    )
107}
108
109/// Returns the last index matching the byte `x` in `text`.
110#[must_use]
111pub fn memrchr(x: u8, text: &[u8]) -> Option<usize> {
112    // Scan for a single byte value by reading two `usize` words at a time.
113    //
114    // Split `text` in three parts:
115    // - unaligned tail, after the last word aligned address in text,
116    // - body, scanned by 2 words at a time,
117    // - the first remaining bytes, < 2 word size.
118    let len = text.len();
119    let ptr = text.as_ptr();
120    type Chunk = usize;
121
122    let (min_aligned_offset, max_aligned_offset) = {
123        // We call this just to obtain the length of the prefix and suffix.
124        // In the middle we always process two chunks at once.
125        // SAFETY: transmuting `[u8]` to `[usize]` is safe except for size differences
126        // which are handled by `align_to`.
127        let (prefix, _, suffix) = unsafe { text.align_to::<(Chunk, Chunk)>() };
128        (prefix.len(), len - suffix.len())
129    };
130
131    let mut offset = max_aligned_offset;
132    if let Some(index) = text[offset..].iter().rposition(|elt| *elt == x) {
133        return Some(offset + index);
134    }
135
136    // Search the body of the text, make sure we don't cross min_aligned_offset.
137    // offset is always aligned, so just testing `>` is sufficient and avoids possible
138    // overflow.
139    let repeated_x = usize::repeat_u8(x);
140    let chunk_bytes = size_of::<Chunk>();
141
142    while offset > min_aligned_offset {
143        // SAFETY: offset starts at len - suffix.len(), as long as it is greater than
144        // min_aligned_offset (prefix.len()) the remaining distance is at least 2 * chunk_bytes.
145        unsafe {
146            let u = *(ptr.add(offset - 2 * chunk_bytes) as *const Chunk);
147            let v = *(ptr.add(offset - chunk_bytes) as *const Chunk);
148
149            // Break if there is a matching byte.
150            let zu = contains_zero_byte(u ^ repeated_x);
151            let zv = contains_zero_byte(v ^ repeated_x);
152            if zu || zv {
153                break;
154            }
155        }
156        offset -= 2 * chunk_bytes;
157    }
158
159    // Find the byte before the point the body loop stopped.
160    text[..offset].iter().rposition(|elt| *elt == x)
161}