vyre 0.4.0

GPU compute intermediate representation with a standard operation library
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

//! BLAKE2s-256 and BLAKE2b-512 CPU reference implementations.

// blake2b_bytes.rs
/// Compute BLAKE2b with caller-selected output length up to 64 bytes.
#[must_use]
pub(crate) fn blake2b_bytes(input: &[u8], out_len: usize) -> Vec<u8> {
    assert!(
        (1..=64).contains(&out_len),
        "BLAKE2b output length must be 1..=64"
    );
    let mut h = IV64;
    h[0] ^= 0x0101_0000 ^ out_len as u64;
    if input.is_empty() {
        compress64(&mut h, &[0; 128], 0, true);
    } else {
        let mut offset = 0usize;
        while offset + 128 < input.len() {
            compress64(
                &mut h,
                &input[offset..offset + 128],
                offset as u128 + 128,
                false,
            );
            offset += 128;
        }
        let mut block = [0u8; 128];
        block[..input.len() - offset].copy_from_slice(&input[offset..]);
        compress64(&mut h, &block, input.len() as u128, true);
    }
    let mut out = Vec::with_capacity(64);
    for word in h {
        out.extend_from_slice(&word.to_le_bytes());
    }
    out.truncate(out_len);
    out
}

// blake2b_words.rs
/// Compute unkeyed BLAKE2b-512 and return eight big-endian digest words.
#[must_use]
pub(crate) fn blake2b_words(input: &[u8]) -> [u64; 8] {
    let mut h = IV64;
    h[0] ^= 0x0101_0040;
    if input.is_empty() {
        compress64(&mut h, &[0; 128], 0, true);
    } else {
        let mut offset = 0usize;
        while offset + 128 < input.len() {
            compress64(
                &mut h,
                &input[offset..offset + 128],
                offset as u128 + 128,
                false,
            );
            offset += 128;
        }
        let mut block = [0u8; 128];
        block[..input.len() - offset].copy_from_slice(&input[offset..]);
        compress64(&mut h, &block, input.len() as u128, true);
    }
    h.map(u64::to_be)
}

// blake2s_words.rs
/// Compute unkeyed BLAKE2s-256 and return eight big-endian digest words.
#[must_use]
pub(crate) fn blake2s_words(input: &[u8]) -> [u32; 8] {
    let mut h = IV32;
    h[0] ^= 0x0101_0020;
    if input.is_empty() {
        compress32(&mut h, &[0; 64], 0, true);
    } else {
        let mut offset = 0usize;
        while offset + 64 < input.len() {
            compress32(
                &mut h,
                &input[offset..offset + 64],
                offset as u64 + 64,
                false,
            );
            offset += 64;
        }
        let mut block = [0u8; 64];
        block[..input.len() - offset].copy_from_slice(&input[offset..]);
        compress32(&mut h, &block, input.len() as u64, true);
    }
    h.map(u32::to_be)
}

// compress32.rs
pub fn compress32(h: &mut [u32; 8], block: &[u8], count: u64, last: bool) {
    let mut m = [0u32; 16];
    for (slot, bytes) in m.iter_mut().zip(block.chunks_exact(4)) {
        *slot = u32::from_le_bytes([bytes[0], bytes[1], bytes[2], bytes[3]]);
    }
    let mut v = [0u32; 16];
    v[..8].copy_from_slice(h);
    v[8..].copy_from_slice(&IV32);
    v[12] ^= count as u32;
    v[13] ^= (count >> 32) as u32;
    if last {
        v[14] = !v[14];
    }
    for s in SIGMA.iter().take(10) {
        g32(&mut v, 0, 4, 8, 12, m[s[0]], m[s[1]]);
        g32(&mut v, 1, 5, 9, 13, m[s[2]], m[s[3]]);
        g32(&mut v, 2, 6, 10, 14, m[s[4]], m[s[5]]);
        g32(&mut v, 3, 7, 11, 15, m[s[6]], m[s[7]]);
        g32(&mut v, 0, 5, 10, 15, m[s[8]], m[s[9]]);
        g32(&mut v, 1, 6, 11, 12, m[s[10]], m[s[11]]);
        g32(&mut v, 2, 7, 8, 13, m[s[12]], m[s[13]]);
        g32(&mut v, 3, 4, 9, 14, m[s[14]], m[s[15]]);
    }
    for i in 0..8 {
        h[i] ^= v[i] ^ v[i + 8];
    }
}

// compress64.rs
pub fn compress64(h: &mut [u64; 8], block: &[u8], count: u128, last: bool) {
    let mut m = [0u64; 16];
    for (slot, bytes) in m.iter_mut().zip(block.chunks_exact(8)) {
        *slot = u64::from_le_bytes([
            bytes[0], bytes[1], bytes[2], bytes[3], bytes[4], bytes[5], bytes[6], bytes[7],
        ]);
    }
    let mut v = [0u64; 16];
    v[..8].copy_from_slice(h);
    v[8..].copy_from_slice(&IV64);
    v[12] ^= count as u64;
    v[13] ^= (count >> 64) as u64;
    if last {
        v[14] = !v[14];
    }
    for s in &SIGMA {
        g64(&mut v, 0, 4, 8, 12, m[s[0]], m[s[1]]);
        g64(&mut v, 1, 5, 9, 13, m[s[2]], m[s[3]]);
        g64(&mut v, 2, 6, 10, 14, m[s[4]], m[s[5]]);
        g64(&mut v, 3, 7, 11, 15, m[s[6]], m[s[7]]);
        g64(&mut v, 0, 5, 10, 15, m[s[8]], m[s[9]]);
        g64(&mut v, 1, 6, 11, 12, m[s[10]], m[s[11]]);
        g64(&mut v, 2, 7, 8, 13, m[s[12]], m[s[13]]);
        g64(&mut v, 3, 4, 9, 14, m[s[14]], m[s[15]]);
    }
    for i in 0..8 {
        h[i] ^= v[i] ^ v[i + 8];
    }
}

// g32.rs
pub fn g32(v: &mut [u32; 16], a: usize, b: usize, c: usize, d: usize, x: u32, y: u32) {
    v[a] = v[a].wrapping_add(v[b]).wrapping_add(x);
    v[d] = (v[d] ^ v[a]).rotate_right(16);
    v[c] = v[c].wrapping_add(v[d]);
    v[b] = (v[b] ^ v[c]).rotate_right(12);
    v[a] = v[a].wrapping_add(v[b]).wrapping_add(y);
    v[d] = (v[d] ^ v[a]).rotate_right(8);
    v[c] = v[c].wrapping_add(v[d]);
    v[b] = (v[b] ^ v[c]).rotate_right(7);
}

// g64.rs
pub fn g64(v: &mut [u64; 16], a: usize, b: usize, c: usize, d: usize, x: u64, y: u64) {
    v[a] = v[a].wrapping_add(v[b]).wrapping_add(x);
    v[d] = (v[d] ^ v[a]).rotate_right(32);
    v[c] = v[c].wrapping_add(v[d]);
    v[b] = (v[b] ^ v[c]).rotate_right(24);
    v[a] = v[a].wrapping_add(v[b]).wrapping_add(y);
    v[d] = (v[d] ^ v[a]).rotate_right(16);
    v[c] = v[c].wrapping_add(v[d]);
    v[b] = (v[b] ^ v[c]).rotate_right(63);
}

// iv32.rs
pub const IV32: [u32; 8] = [
    0x6a09e667, 0xbb67ae85, 0x3c6ef372, 0xa54ff53a, 0x510e527f, 0x9b05688c, 0x1f83d9ab, 0x5be0cd19,
];

// iv64.rs
pub const IV64: [u64; 8] = [
    0x6a09e667f3bcc908,
    0xbb67ae8584caa73b,
    0x3c6ef372fe94f82b,
    0xa54ff53a5f1d36f1,
    0x510e527fade682d1,
    0x9b05688c2b3e6c1f,
    0x1f83d9abfb41bd6b,
    0x5be0cd19137e2179,
];

// sigma.rs
pub const SIGMA: [[usize; 16]; 12] = [
    [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
    [14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3],
    [11, 8, 12, 0, 5, 2, 15, 13, 10, 14, 3, 6, 7, 1, 9, 4],
    [7, 9, 3, 1, 13, 12, 11, 14, 2, 6, 5, 10, 4, 0, 15, 8],
    [9, 0, 5, 7, 2, 4, 10, 15, 14, 1, 11, 12, 6, 8, 3, 13],
    [2, 12, 6, 10, 0, 11, 8, 3, 4, 13, 7, 5, 15, 14, 1, 9],
    [12, 5, 1, 15, 14, 13, 4, 10, 0, 7, 6, 3, 9, 2, 8, 11],
    [13, 11, 7, 14, 12, 1, 3, 9, 5, 0, 15, 4, 8, 6, 2, 10],
    [6, 15, 14, 9, 11, 3, 0, 8, 12, 2, 13, 7, 1, 4, 10, 5],
    [10, 2, 8, 4, 7, 6, 1, 5, 15, 11, 9, 14, 3, 12, 13, 0],
    [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
    [14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3],
];