1 // Original implementation taken from rust-memchr.
2 // Copyright 2015 Andrew Gallant, bluss and Nicolas Koch
7 const LO_U64: u64 = 0x0101010101010101;
8 const HI_U64: u64 = 0x8080808080808080;
11 const LO_USIZE: usize = LO_U64 as usize;
12 const HI_USIZE: usize = HI_U64 as usize;
13 const USIZE_BYTES: usize = mem::size_of::<usize>();
15 /// Returns `true` if `x` contains any zero byte.
17 /// From *Matters Computational*, J. Arndt:
19 /// "The idea is to subtract one from each of the bytes and then look for
20 /// bytes where the borrow propagated all the way to the most significant
23 fn contains_zero_byte(x: usize) -> bool {
24 x.wrapping_sub(LO_USIZE) & !x & HI_USIZE != 0
27 #[cfg(target_pointer_width = "16")]
29 fn repeat_byte(b: u8) -> usize {
30 (b as usize) << 8 | b as usize
33 #[cfg(not(target_pointer_width = "16"))]
35 fn repeat_byte(b: u8) -> usize {
36 (b as usize) * (usize::MAX / 255)
39 /// Returns the first index matching the byte `x` in `text`.
42 pub fn memchr(x: u8, text: &[u8]) -> Option<usize> {
43 // Fast path for small slices
44 if text.len() < 2 * USIZE_BYTES {
45 return text.iter().position(|elt| *elt == x);
48 memchr_general_case(x, text)
51 fn memchr_general_case(x: u8, text: &[u8]) -> Option<usize> {
52 // Scan for a single byte value by reading two `usize` words at a time.
54 // Split `text` in three parts
55 // - unaligned initial part, before the first word aligned address in text
56 // - body, scan by 2 words at a time
57 // - the last remaining part, < 2 word size
59 // search up to an aligned boundary
61 let ptr = text.as_ptr();
62 let mut offset = ptr.align_offset(USIZE_BYTES);
65 offset = cmp::min(offset, len);
66 if let Some(index) = text[..offset].iter().position(|elt| *elt == x) {
71 // search the body of the text
72 let repeated_x = repeat_byte(x);
73 while offset <= len - 2 * USIZE_BYTES {
74 // SAFETY: the while's predicate guarantees a distance of at least 2 * usize_bytes
75 // between the offset and the end of the slice.
77 let u = *(ptr.add(offset) as *const usize);
78 let v = *(ptr.add(offset + USIZE_BYTES) as *const usize);
80 // break if there is a matching byte
81 let zu = contains_zero_byte(u ^ repeated_x);
82 let zv = contains_zero_byte(v ^ repeated_x);
87 offset += USIZE_BYTES * 2;
90 // Find the byte after the point the body loop stopped.
91 text[offset..].iter().position(|elt| *elt == x).map(|i| offset + i)
94 /// Returns the last index matching the byte `x` in `text`.
96 pub fn memrchr(x: u8, text: &[u8]) -> Option<usize> {
97 // Scan for a single byte value by reading two `usize` words at a time.
99 // Split `text` in three parts:
100 // - unaligned tail, after the last word aligned address in text,
101 // - body, scanned by 2 words at a time,
102 // - the first remaining bytes, < 2 word size.
103 let len = text.len();
104 let ptr = text.as_ptr();
107 let (min_aligned_offset, max_aligned_offset) = {
108 // We call this just to obtain the length of the prefix and suffix.
109 // In the middle we always process two chunks at once.
110 // SAFETY: transmuting `[u8]` to `[usize]` is safe except for size differences
111 // which are handled by `align_to`.
112 let (prefix, _, suffix) = unsafe { text.align_to::<(Chunk, Chunk)>() };
113 (prefix.len(), len - suffix.len())
116 let mut offset = max_aligned_offset;
117 if let Some(index) = text[offset..].iter().rposition(|elt| *elt == x) {
118 return Some(offset + index);
121 // Search the body of the text, make sure we don't cross min_aligned_offset.
122 // offset is always aligned, so just testing `>` is sufficient and avoids possible
124 let repeated_x = repeat_byte(x);
125 let chunk_bytes = mem::size_of::<Chunk>();
127 while offset > min_aligned_offset {
128 // SAFETY: offset starts at len - suffix.len(), as long as it is greater than
129 // min_aligned_offset (prefix.len()) the remaining distance is at least 2 * chunk_bytes.
131 let u = *(ptr.offset(offset as isize - 2 * chunk_bytes as isize) as *const Chunk);
132 let v = *(ptr.offset(offset as isize - chunk_bytes as isize) as *const Chunk);
134 // Break if there is a matching byte.
135 let zu = contains_zero_byte(u ^ repeated_x);
136 let zv = contains_zero_byte(v ^ repeated_x);
141 offset -= 2 * chunk_bytes;
144 // Find the byte before the point the body loop stopped.
145 text[..offset].iter().rposition(|elt| *elt == x)