itoap 1.0.1

Even faster functions for printing integers with decimal format
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

#![allow(non_upper_case_globals)]

#[cfg(target_arch = "x86")]
use core::arch::x86::*;
#[cfg(target_arch = "x86_64")]
use core::arch::x86_64::*;

use crate::common::{divmod, lookup, write4, write4_pad};
use core::ptr;

#[repr(align(16))]
struct Aligned<T>(T);

impl<T> std::ops::Deref for Aligned<T> {
    type Target = T;

    #[inline]
    fn deref(&self) -> &T {
        &self.0
    }
}

const kDiv10000: u32 = 0xd1b71759;
const kDivPowersVector: Aligned<[u16; 8]> =
    Aligned([8389, 5243, 13108, 32768, 8389, 5243, 13108, 32768]);
const kShiftPowersVector: Aligned<[u16; 8]> = Aligned([
    1 << (16 - (23 + 2 - 16)),
    1 << (16 - (19 + 2 - 16)),
    1 << (16 - 1 - 2),
    1 << (15),
    1 << (16 - (23 + 2 - 16)),
    1 << (16 - (19 + 2 - 16)),
    1 << (16 - 1 - 2),
    1 << (15),
]);

#[inline]
unsafe fn convert_8digits_sse2(value: u32) -> __m128i {
    debug_assert!(value <= 99999999);

    // abcd, efgh = abcdefgh divmod 10000
    let abcdefgh = _mm_cvtsi32_si128(value as i32);
    let abcd = _mm_srli_epi64(
        _mm_mul_epu32(abcdefgh, _mm_set1_epi32(kDiv10000 as i32)),
        45,
    );
    let efgh = _mm_sub_epi32(abcdefgh, _mm_mul_epu32(abcd, _mm_set1_epi32(10000)));

    // v1 = [ abcd, efgh, 0, 0, 0, 0, 0, 0 ]
    let v1 = _mm_unpacklo_epi16(abcd, efgh);

    // v1a = v1 * 4 = [ abcd*4, efgh*4, 0, 0, 0, 0, 0, 0 ]
    let v1a = _mm_slli_epi64(v1, 2);

    // v2 = [abcd*4, abcd*4, abcd*4, abcd*4, efgh*4, efgh*4, efgh*4, efgh*4]
    let v2a = _mm_unpacklo_epi16(v1a, v1a);
    let v2 = _mm_unpacklo_epi32(v2a, v2a);

    // v4 = v2 div 10^3, 10^2, 10^1, 10^0 = [ a, ab, abc, abcd, e, ef, efg, efgh ]
    let v3 = _mm_mulhi_epu16(
        v2,
        _mm_load_si128(kDivPowersVector.as_ptr() as *const __m128i),
    );
    let v4 = _mm_mulhi_epu16(
        v3,
        _mm_load_si128(kShiftPowersVector.as_ptr() as *const __m128i),
    );

    // v5 = v4 * 10 = [ a0, ab0, abc0, abcd0, e0, ef0, efg0, efgh0 ]
    let v5 = _mm_mullo_epi16(v4, _mm_set1_epi16(10));

    // v6 = v5 << 16 = [ 0, a0, ab0, abc0, 0, e0, ef0, efg0 ]
    let v6 = _mm_slli_epi64(v5, 16);

    // v4 - v6 = { a, b, c, d, e, f, g, h }
    _mm_sub_epi16(v4, v6)
}

pub unsafe fn write_u32(n: u32, buf: *mut u8) -> usize {
    if n < 10000 {
        write4(n, buf)
    } else if n < 100_000_000 {
        let (n1, n2) = divmod(n, 10000);

        let l = write4(n1, buf);
        write4_pad(n2, buf.add(l));
        l + 4
    } else {
        let (n1, n2) = divmod(n, 100_000_000);

        let l = if n1 >= 10 {
            ptr::copy_nonoverlapping(lookup(n1), buf, 2);
            2
        } else {
            *buf = n1 as u8 + 0x30;
            1
        };

        let b = convert_8digits_sse2(n2);
        let ba = _mm_add_epi8(
            _mm_packus_epi16(_mm_setzero_si128(), b),
            _mm_set1_epi8(b'0' as i8),
        );
        let result = _mm_srli_si128(ba, 8);
        _mm_storel_epi64(buf.add(l) as *mut __m128i, result);

        l + 8
    }
}

pub unsafe fn write_u64(n: u64, buf: *mut u8) -> usize {
    if n < 10000 {
        write4(n as u32, buf)
    } else if n < 100_000_000 {
        let (n1, n2) = divmod(n as u32, 10000);

        let l = write4(n1, buf);
        write4_pad(n2, buf.add(l));
        l + 4
    } else if n < 10_000_000_000_000_000 {
        let (n1, n2) = divmod(n, 100_000_000);
        let (n1, n2) = (n1 as u32, n2 as u32);

        let l = if n1 < 10000 {
            write4(n1, buf)
        } else {
            let (n11, n12) = divmod(n1, 10000);
            let l = write4(n11, buf);
            write4_pad(n12, buf.add(l));
            l + 4
        };

        let b = convert_8digits_sse2(n2);
        let ba = _mm_add_epi8(
            _mm_packus_epi16(_mm_setzero_si128(), b),
            _mm_set1_epi8(b'0' as i8),
        );
        let result = _mm_srli_si128(ba, 8);
        _mm_storel_epi64(buf.add(l) as *mut __m128i, result);

        l + 8
    } else {
        let (n1, n2) = divmod(n, 10_000_000_000_000_000);
        let l = write4(n1 as u32, buf);

        let (n21, n22) = divmod(n2, 100_000_000);

        let a0 = convert_8digits_sse2(n21 as u32);
        let a1 = convert_8digits_sse2(n22 as u32);

        // Convert to bytes, add '0'
        let va = _mm_add_epi8(_mm_packus_epi16(a0, a1), _mm_set1_epi8(b'0' as i8));
        _mm_storeu_si128(buf.add(l) as *mut __m128i, va);

        l + 16
    }
}