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

use simd_abstraction::traits::SIMD256;

#[allow(missing_docs, clippy::missing_safety_doc)]
pub unsafe trait SIMDExt: SIMD256 {
    fn base64_split_bits(self, a: Self::V256) -> Self::V256 {
        // a : {bbbbcccc|aaaaaabb|ccdddddd|bbbbcccc} x8 (1021)

        let m1 = self.u32x8_splat(u32::from_le_bytes([0x00, 0xfc, 0x00, 0x00]));
        let a1 = self.u16x16_shr::<10>(self.v256_and(a, m1));
        // a1: {00aaaaaa|000000000|00000000|00000000} x8

        let m2 = self.u32x8_splat(u32::from_le_bytes([0xf0, 0x03, 0x00, 0x00]));
        let a2 = self.u16x16_shl::<4>(self.v256_and(a, m2));
        // a2: {00000000|00bbbbbb|00000000|00000000} x8

        let m3 = self.u32x8_splat(u32::from_le_bytes([0x00, 0x00, 0xc0, 0x0f]));
        let a3 = self.u16x16_shr::<6>(self.v256_and(a, m3));
        // a3: {00000000|00000000|00cccccc|00000000} x8

        let m4 = self.u32x8_splat(u32::from_le_bytes([0x00, 0x00, 0x3f, 0x00]));
        let a4 = self.u16x16_shl::<8>(self.v256_and(a, m4));
        // a4: {00000000|00000000|00000000|00dddddd} x8

        self.v256_or(self.v256_or(a1, a2), self.v256_or(a3, a4))
        // {00aaaaaa|00bbbbbb|00cccccc|00dddddd} x8
    }

    fn base64_merge_bits(self, a: Self::V256) -> Self::V256 {
        // a : {00aaaaaa|00bbbbbb|00cccccc|00dddddd} x8

        let m1 = self.u32x8_splat(u32::from_le_bytes([0x3f, 0x00, 0x3f, 0x00]));
        let a1 = self.v256_and(a, m1);
        // a1: {00aaaaaa|00000000|00cccccc|00000000} x8

        let m2 = self.u32x8_splat(u32::from_le_bytes([0x00, 0x3f, 0x00, 0x3f]));
        let a2 = self.v256_and(a, m2);
        // a2: {00000000|00bbbbbb|00000000|00dddddd} x8

        let a3 = self.v256_or(self.u32x8_shl::<18>(a1), self.u32x8_shr::<10>(a1));
        // a3: {cc000000|0000cccc|aaaaaa00|00000000} x8

        let a4 = self.v256_or(self.u32x8_shl::<4>(a2), self.u32x8_shr::<24>(a2));
        // a4: {00dddddd|bbbb0000|000000bb|dddd0000}

        let mask = self.u32x8_splat(u32::from_le_bytes([0xff, 0xff, 0xff, 0x00]));
        self.v256_and(self.v256_or(a3, a4), mask)
        // {ccdddddd|bbbbcccc|aaaaaabb|00000000} x8
    }
}

#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
mod x86 {
    use super::SIMDExt;

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

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

    use simd_abstraction::arch::x86::*;
    use simd_abstraction::traits::{SIMD128, SIMD256};

    const SPLIT_M1: u32 = u32::from_le_bytes([0x00, 0xfc, 0xc0, 0x0f]);
    const SPLIT_M2: u32 = u32::from_le_bytes([0xf0, 0x03, 0x3f, 0x00]);
    const SPLIT_M3: u32 = u32::from_le_bytes([0x40, 0x00, 0x00, 0x04]);
    const SPLIT_M4: u32 = u32::from_le_bytes([0x10, 0x00, 0x00, 0x01]);

    const MERGE_M1: u16 = u16::from_le_bytes([0x40, 0x01]);
    const MERGE_M2: u32 = u32::from_le_bytes([0x00, 0x10, 0x01, 0x00]);

    unsafe impl SIMDExt for AVX2 {
        #[inline(always)]
        fn base64_split_bits(self, a: Self::V256) -> Self::V256 {
            // a : {bbbbcccc|aaaaaabb|ccdddddd|bbbbcccc} x8 (1021)

            let a1 = self.v256_and(a, self.u32x8_splat(SPLIT_M1));
            // a1: {00000000|aaaaaa00|cc000000|0000cccc} x8

            let a2 = self.v256_and(a, self.u32x8_splat(SPLIT_M2));
            // a2: {bbbb0000|000000bb|00dddddd|00000000} x8

            let a3 = unsafe { _mm256_mulhi_epu16(a1, self.u32x8_splat(SPLIT_M3)) };
            // a3: {00aaaaaa|00000000|00cccccc|00000000} x8

            let a4 = unsafe { _mm256_mullo_epi16(a2, self.u32x8_splat(SPLIT_M4)) };
            // a4: {00000000|00bbbbbb|00000000|00dddddd} x8

            self.v256_or(a3, a4)
            // {00aaaaaa|00bbbbbb|00cccccc|00dddddd} x8
        }

        #[inline(always)]
        fn base64_merge_bits(self, a: Self::V256) -> Self::V256 {
            // a : {00aaaaaa|00bbbbbb|00cccccc|00dddddd} x8

            let a1 = unsafe { _mm256_maddubs_epi16(a, self.u16x16_splat(MERGE_M1)) };
            // a1: {aabbbbbb|0000aaaa|ccdddddd|0000cccc} x8

            unsafe { _mm256_madd_epi16(a1, self.u32x8_splat(MERGE_M2)) }
            // {00000000|aaaaaabb|bbbbcccc|ccdddddd} x8
        }
    }

    unsafe impl SIMDExt for SSE41 {
        #[inline(always)]
        fn base64_split_bits(self, a: Self::V256) -> Self::V256 {
            // a : {bbbbcccc|aaaaaabb|ccdddddd|bbbbcccc} x8 (1021)

            let a1 = self.v256_and(a, self.u32x8_splat(SPLIT_M1));
            // a1: {00000000|aaaaaa00|cc000000|0000cccc} x8

            let a2 = self.v256_and(a, self.u32x8_splat(SPLIT_M2));
            // a2: {bbbb0000|000000bb|00dddddd|00000000} x8

            let a3 = unsafe {
                let m3 = self.u32x4_splat(SPLIT_M3);
                (_mm_mulhi_epu16(a1.0, m3), _mm_mulhi_epu16(a1.1, m3))
            };
            // a3: {00aaaaaa|00000000|00cccccc|00000000} x8

            let a4 = unsafe {
                let m4 = self.u32x4_splat(SPLIT_M4);
                (_mm_mullo_epi16(a2.0, m4), _mm_mullo_epi16(a2.1, m4))
            };
            // a4: {00000000|00bbbbbb|00000000|00dddddd} x8

            self.v256_or(a3, a4)
            // {00aaaaaa|00bbbbbb|00cccccc|00dddddd} x8
        }

        #[inline(always)]
        fn base64_merge_bits(self, a: Self::V256) -> Self::V256 {
            // a : {00aaaaaa|00bbbbbb|00cccccc|00dddddd} x8

            let a1 = unsafe {
                let m1 = self.u16x8_splat(MERGE_M1);
                (_mm_maddubs_epi16(a.0, m1), _mm_maddubs_epi16(a.1, m1))
            };
            // a1: {aabbbbbb|0000aaaa|ccdddddd|0000cccc} x8

            unsafe {
                let m2 = self.u32x4_splat(MERGE_M2);
                (_mm_madd_epi16(a1.0, m2), _mm_madd_epi16(a1.1, m2))
            }
            // {ccdddddd|bbbbcccc|aaaaaabb|00000000} x8
        }
    }
}

#[cfg(target_arch = "wasm32")]
mod wasm {
    use super::SIMDExt;

    use simd_abstraction::arch::wasm::*;

    unsafe impl SIMDExt for SIMD128 {}
}

#[cfg(all(
    feature = "unstable",
    any(target_arch = "arm", target_arch = "aarch64")
))]
mod arm {
    use super::SIMDExt;

    #[cfg(target_arch = "arm")]
    use simd_abstraction::arch::arm::*;

    #[cfg(target_arch = "aarch64")]
    use simd_abstraction::arch::aarch64::*;

    unsafe impl SIMDExt for NEON {} // TODO: better ways?
}