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
|
extern crate packed_simd;
#[cfg(not(feature = "runtime-dispatch-simd"))]
use core::mem;
#[cfg(feature = "runtime-dispatch-simd")]
use std::mem;
use self::packed_simd::{u8x32, u8x64, FromCast};
const MASK: [u8; 64] = [
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
];
unsafe fn u8x64_from_offset(slice: &[u8], offset: usize) -> u8x64 {
u8x64::from_slice_unaligned_unchecked(slice.get_unchecked(offset..))
}
unsafe fn u8x32_from_offset(slice: &[u8], offset: usize) -> u8x32 {
u8x32::from_slice_unaligned_unchecked(slice.get_unchecked(offset..))
}
fn sum_x64(u8s: &u8x64) -> usize {
let mut store = [0; mem::size_of::<u8x64>()];
u8s.write_to_slice_unaligned(&mut store);
store.iter().map(|&e| e as usize).sum()
}
fn sum_x32(u8s: &u8x32) -> usize {
let mut store = [0; mem::size_of::<u8x32>()];
u8s.write_to_slice_unaligned(&mut store);
store.iter().map(|&e| e as usize).sum()
}
pub fn chunk_count(haystack: &[u8], needle: u8) -> usize {
assert!(haystack.len() >= 32);
unsafe {
let mut offset = 0;
let mut count = 0;
let needles_x64 = u8x64::splat(needle);
// 16320
while haystack.len() >= offset + 64 * 255 {
let mut counts = u8x64::splat(0);
for _ in 0..255 {
counts -= u8x64::from_cast(u8x64_from_offset(haystack, offset).eq(needles_x64));
offset += 64;
}
count += sum_x64(&counts);
}
// 8192
if haystack.len() >= offset + 64 * 128 {
let mut counts = u8x64::splat(0);
for _ in 0..128 {
counts -= u8x64::from_cast(u8x64_from_offset(haystack, offset).eq(needles_x64));
offset += 64;
}
count += sum_x64(&counts);
}
let needles_x32 = u8x32::splat(needle);
// 32
let mut counts = u8x32::splat(0);
for i in 0..(haystack.len() - offset) / 32 {
counts -= u8x32::from_cast(u8x32_from_offset(haystack, offset + i * 32).eq(needles_x32));
}
count += sum_x32(&counts);
// Straggler; need to reset counts because prior loop can run 255 times
counts = u8x32::splat(0);
if haystack.len() % 32 != 0 {
counts -= u8x32::from_cast(u8x32_from_offset(haystack, haystack.len() - 32).eq(needles_x32)) &
u8x32_from_offset(&MASK, haystack.len() % 32);
}
count += sum_x32(&counts);
count
}
}
fn is_leading_utf8_byte_x64(u8s: u8x64) -> u8x64 {
u8x64::from_cast((u8s & u8x64::splat(0b1100_0000)).ne(u8x64::splat(0b1000_0000)))
}
fn is_leading_utf8_byte_x32(u8s: u8x32) -> u8x32 {
u8x32::from_cast((u8s & u8x32::splat(0b1100_0000)).ne(u8x32::splat(0b1000_0000)))
}
pub fn chunk_num_chars(utf8_chars: &[u8]) -> usize {
assert!(utf8_chars.len() >= 32);
unsafe {
let mut offset = 0;
let mut count = 0;
// 16320
while utf8_chars.len() >= offset + 64 * 255 {
let mut counts = u8x64::splat(0);
for _ in 0..255 {
counts -= is_leading_utf8_byte_x64(u8x64_from_offset(utf8_chars, offset));
offset += 64;
}
count += sum_x64(&counts);
}
// 8192
if utf8_chars.len() >= offset + 64 * 128 {
let mut counts = u8x64::splat(0);
for _ in 0..128 {
counts -= is_leading_utf8_byte_x64(u8x64_from_offset(utf8_chars, offset));
offset += 64;
}
count += sum_x64(&counts);
}
// 32
let mut counts = u8x32::splat(0);
for i in 0..(utf8_chars.len() - offset) / 32 {
counts -= is_leading_utf8_byte_x32(u8x32_from_offset(utf8_chars, offset + i * 32));
}
count += sum_x32(&counts);
// Straggler; need to reset counts because prior loop can run 255 times
counts = u8x32::splat(0);
if utf8_chars.len() % 32 != 0 {
counts -= is_leading_utf8_byte_x32(u8x32_from_offset(utf8_chars, utf8_chars.len() - 32)) &
u8x32_from_offset(&MASK, utf8_chars.len() % 32);
}
count += sum_x32(&counts);
count
}
}
|
