radix64 0.6.2

A base64 library by another name
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
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247

/// SSE implementation of base64 encoding.
use crate::Config;
use crate::encode::block::{BlockEncoder, IntoBlockEncoder, ScalarBlockEncoder};
use crate::{Std, StdNoPad, UrlSafe, UrlSafeNoPad, Crypt, Fast};

#[derive(Debug,Clone,Copy)]
pub struct Encoder<C>(C);

impl<C> BlockEncoder for Encoder<C> where C: Config + avx2::Translate256i {
    #[inline]
    fn encode_blocks(self, input: &[u8], output: &mut [u8]) -> (usize, usize) {
        if let Ok(encoder) = avx2::Encoder::new(self.0) {
            encoder.encode_blocks(input, output)
        } else {
            ScalarBlockEncoder::new(self.0).encode_blocks(input, output)
        }
    }
}

macro_rules! define_into_block_encoder {
    ($( $cfg:ident ),+) => {$(
        impl IntoBlockEncoder for $cfg {
            type BlockEncoder = Encoder<Self>;

            #[inline]
            fn into_block_encoder(self) -> Self::BlockEncoder {
                Encoder(self)
            }
        }
    )+}
}
define_into_block_encoder!(Std,StdNoPad,UrlSafe,UrlSafeNoPad,Crypt,Fast);

mod avx2 {
     #[cfg(target_arch = "x86")]
    use std::arch::x86::*;
    #[cfg(target_arch = "x86_64")]
    use std::arch::x86_64::*;
    use crate::{Std, StdNoPad, UrlSafe, UrlSafeNoPad, Crypt, Fast};

    pub trait Translate256i: Copy {
        unsafe fn translate_m256i(input: __m256i) -> __m256i;
    }

    #[derive(Debug,Clone,Copy)]
    pub(crate) struct Encoder<C>(C);

    impl<C> Encoder<C> where C: Translate256i {
        #[inline]
        pub(crate) fn new(config: C) -> Result<Self, ()> {
            if is_x86_feature_detected!("avx2") {
                Ok(Encoder(config))
            } else {
                Err(())
            }
        }

        pub(crate) fn encode_blocks(self, input: &[u8], output: &mut [u8]) -> (usize, usize) {
            // The unsafe block is required because _encode_blocks relies on AVX2
            // intrinsics. This is safe because Encoder::new() ensures that an
            // encoder is only successfully created when the CPU supports AVX2.
            unsafe { self._encode_blocks(input, output) }
        }

        #[target_feature(enable = "avx2")]
        unsafe fn _encode_blocks(self, input: &[u8], output: &mut [u8]) -> (usize, usize) {
            let mut iter = BlockIter::new(input, output);
            while let Some((input, output)) = iter.next_chunk() {
                #[allow(clippy::cast_ptr_alignment)]
                let lo_data = _mm_loadu_si128(input.as_ptr() as *const __m128i);
                #[allow(clippy::cast_ptr_alignment)]
                let hi_data = _mm_loadu_si128(input.as_ptr().add(12) as *const __m128i);
                let input = _mm256_set_m128i(hi_data, lo_data);
                #[allow(clippy::cast_ptr_alignment)]
                _mm256_storeu_si256(output.as_mut_ptr() as *mut __m256i, self.encode_block(input));
            }
            iter.remaining()
        }

        #[target_feature(enable = "avx2")]
        unsafe fn encode_block(self, input: __m256i) -> __m256i {
            #[rustfmt::skip]
            let input = _mm256_shuffle_epi8(
                input,
                _mm256_setr_epi8(
                    2,  2,  1,  0,  // The trailing comments fix a bug in tarpaulin
                    5,  5,  4,  3,  // causing the args to be lines not covered.
                    8,  8,  7,  6,  //
                    11, 11, 10, 9,  //
                    2,  2,  1,  0,  //
                    5,  5,  4,  3,  //
                    8,  8,  7,  6,  //
                    11, 11, 10, 9,  //
                ),
            );
            let mask = _mm256_set1_epi32(0x3F00_0000);
            let res = _mm256_and_si256(_mm256_srli_epi32(input, 2), mask);
            let mask = _mm256_srli_epi32(mask, 8);
            let res = _mm256_or_si256(res, _mm256_and_si256(_mm256_srli_epi32(input, 4), mask));
            let mask = _mm256_srli_epi32(mask, 8);
            let res = _mm256_or_si256(res, _mm256_and_si256(_mm256_srli_epi32(input, 6), mask));
            let mask = _mm256_srli_epi32(mask, 8);
            let res = _mm256_or_si256(res, _mm256_and_si256(input, mask));
            #[rustfmt::skip]
            let res = _mm256_shuffle_epi8(
                res,
                _mm256_setr_epi8(
                    3,  2,  1,  0,  // The trailing comments fix a bug in tarpaulin
                    7,  6,  5,  4,  // causing the args to be lines not covered.
                    11, 10, 9,  8,  //
                    15, 14, 13, 12, //
                    19, 18, 17, 16, //
                    23, 22, 21, 20, //
                    27, 26, 25, 24, //
                    31, 30, 29, 28, //
                ),
            );
            C::translate_m256i(res)
        }

    }

    define_block_iter!(name=BlockIter, input_chunk_size=28, input_stride=24, output_chunk_size=32, output_stride=32);

    #[target_feature(enable = "avx2")]
    #[inline]
    unsafe fn translate_std(input: __m256i) -> __m256i {
        let s1mask = _mm256_cmpgt_epi8(_mm256_set1_epi8(26), input);
        let mut blockmask = s1mask;
        let s2mask = _mm256_andnot_si256(blockmask, _mm256_cmpgt_epi8(_mm256_set1_epi8(52), input));
        blockmask = _mm256_or_si256(blockmask, s2mask);
        let s3mask = _mm256_andnot_si256(blockmask, _mm256_cmpgt_epi8(_mm256_set1_epi8(62), input));
        blockmask = _mm256_or_si256(blockmask, s3mask);
        let s4mask = _mm256_andnot_si256(blockmask, _mm256_cmpgt_epi8(_mm256_set1_epi8(63), input));
        blockmask = _mm256_or_si256(blockmask, s4mask);
        let s1 = _mm256_and_si256(s1mask, _mm256_add_epi8(input, _mm256_set1_epi8(b'A' as i8)));
        let s2 = _mm256_and_si256(
            s2mask,
            _mm256_add_epi8(input, _mm256_set1_epi8(b'a' as i8 - 26)),
        );
        let s3 = _mm256_and_si256(
            s3mask,
            _mm256_add_epi8(input, _mm256_set1_epi8(b'0' as i8 - 52)),
        );
        let s4 = _mm256_and_si256(s4mask, _mm256_set1_epi8(b'+' as i8));
        let s5 = _mm256_andnot_si256(blockmask, _mm256_set1_epi8(b'/' as i8));
        _mm256_or_si256(
            s1,
            _mm256_or_si256(s2, _mm256_or_si256(s3, _mm256_or_si256(s4, s5))),
        )
    }

    #[target_feature(enable = "avx2")]
    #[inline]
    unsafe fn translate_url_safe(input: __m256i) -> __m256i {
        let s1mask = _mm256_cmpgt_epi8(_mm256_set1_epi8(26), input);
        let mut blockmask = s1mask;
        let s2mask = _mm256_andnot_si256(blockmask, _mm256_cmpgt_epi8(_mm256_set1_epi8(52), input));
        blockmask = _mm256_or_si256(blockmask, s2mask);
        let s3mask = _mm256_andnot_si256(blockmask, _mm256_cmpgt_epi8(_mm256_set1_epi8(62), input));
        blockmask = _mm256_or_si256(blockmask, s3mask);
        let s4mask = _mm256_andnot_si256(blockmask, _mm256_cmpgt_epi8(_mm256_set1_epi8(63), input));
        blockmask = _mm256_or_si256(blockmask, s4mask);
        let s1 = _mm256_and_si256(s1mask, _mm256_add_epi8(input, _mm256_set1_epi8(b'A' as i8)));
        let s2 = _mm256_and_si256(
            s2mask,
            _mm256_add_epi8(input, _mm256_set1_epi8(b'a' as i8 - 26)),
        );
        let s3 = _mm256_and_si256(
            s3mask,
            _mm256_add_epi8(input, _mm256_set1_epi8(b'0' as i8 - 52)),
        );
        let s4 = _mm256_and_si256(s4mask, _mm256_set1_epi8(b'-' as i8));
        let s5 = _mm256_andnot_si256(blockmask, _mm256_set1_epi8(b'_' as i8));
        _mm256_or_si256(
            s1,
            _mm256_or_si256(s2, _mm256_or_si256(s3, _mm256_or_si256(s4, s5))),
        )
    }

    #[target_feature(enable = "avx2")]
    #[inline]
    unsafe fn translate_crypt(input: __m256i) -> __m256i {
        let s1mask = _mm256_cmpgt_epi8(_mm256_set1_epi8(12), input);
        let mut blockmask = s1mask;
        let s2mask = _mm256_andnot_si256(blockmask, _mm256_cmpgt_epi8(_mm256_set1_epi8(38), input));
        blockmask = _mm256_or_si256(blockmask, s2mask);
        let s1 = _mm256_and_si256(s1mask, _mm256_add_epi8(input, _mm256_set1_epi8(b'.' as i8)));
        let s2 = _mm256_and_si256(
            s2mask,
            _mm256_add_epi8(input, _mm256_set1_epi8(b'A' as i8 - 12)),
        );
        let s3 = _mm256_andnot_si256(
            blockmask,
            _mm256_add_epi8(input, _mm256_set1_epi8(b'a' as i8 - 38)),
        );
        _mm256_or_si256(s1, _mm256_or_si256(s2, s3))
    }

    #[target_feature(enable = "avx2")]
    #[inline]
    unsafe fn translate_fast(input: __m256i) -> __m256i {
        _mm256_add_epi8(input, _mm256_set1_epi8(62))
    }

    impl Translate256i for Std {
        #[inline]
        unsafe fn translate_m256i(input: __m256i) -> __m256i {
            translate_std(input)
        }
    }

    impl Translate256i for StdNoPad {
        #[inline]
        unsafe fn translate_m256i(input: __m256i) -> __m256i {
            translate_std(input)
        }
    }

    impl Translate256i for UrlSafe {
        #[inline]
        unsafe fn translate_m256i(input: __m256i) -> __m256i {
            translate_url_safe(input)
        }
    }

    impl Translate256i for UrlSafeNoPad {
        #[inline]
        unsafe fn translate_m256i(input: __m256i) -> __m256i {
            translate_url_safe(input)
        }
    }

    impl Translate256i for Crypt {
        #[inline]
        unsafe fn translate_m256i(input: __m256i) -> __m256i {
            translate_crypt(input)
        }
    }

    impl Translate256i for Fast {
        #[inline]
        unsafe fn translate_m256i(input: __m256i) -> __m256i {
            translate_fast(input)
        }
    }
}