libcrux-sha3 0.0.2-alpha.2

Libcrux SHA-3 implementation
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

use libcrux_intrinsics::arm64::*;

use crate::traits::internal::KeccakItem;

#[allow(non_camel_case_types)]
pub type uint64x2_t = _uint64x2_t;

// This file optimizes for the stable Rust Neon Intrinsics
// If we want to use the unstable neon-sha3 instructions, we could use:
// veor3q_u64, vrax1q_u64, vxarq_u64, and vbcaxq_u64
// These instructions might speed up our code even more.

#[inline(always)]
fn rotate_left<const LEFT: i32, const RIGHT: i32>(x: uint64x2_t) -> uint64x2_t {
    debug_assert!(LEFT + RIGHT == 64);
    // The following looks faster but is actually significantly slower
    //unsafe { vsriq_n_u64::<RIGHT>(vshlq_n_u64::<LEFT>(x), x) }
    _veorq_u64(_vshlq_n_u64::<LEFT>(x), _vshrq_n_u64::<RIGHT>(x))
}

#[inline(always)]
fn _veor5q_u64(
    a: uint64x2_t,
    b: uint64x2_t,
    c: uint64x2_t,
    d: uint64x2_t,
    e: uint64x2_t,
) -> uint64x2_t {
    let ab = _veorq_u64(a, b);
    let cd = _veorq_u64(c, d);
    let abcd = _veorq_u64(ab, cd);
    _veorq_u64(abcd, e)
    // Needs nightly+neon-sha3
    //unsafe {veor3q_u64(veor3q_u64(a,b,c),d,e)}
}

#[inline(always)]
fn _vrax1q_u64(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
    _veorq_u64(a, rotate_left::<1, 63>(b))
    // Needs nightly+neon-sha3
    //unsafe { vrax1q_u64(a, b) }
}

#[inline(always)]
fn _vxarq_u64<const LEFT: i32, const RIGHT: i32>(a: uint64x2_t, b: uint64x2_t) -> uint64x2_t {
    let ab = _veorq_u64(a, b);
    rotate_left::<LEFT, RIGHT>(ab)
    // Needs nightly+neon-sha3
    // unsafe { vxarq_u64::<RIGHT>(a,b) }
}

#[inline(always)]
fn _vbcaxq_u64(a: uint64x2_t, b: uint64x2_t, c: uint64x2_t) -> uint64x2_t {
    _veorq_u64(a, _vbicq_u64(b, c))
    // Needs nightly+neon-sha3
    // unsafe{ vbcaxq_u64(a, b, c) }
}

#[inline(always)]
fn _veorq_n_u64(a: uint64x2_t, c: u64) -> uint64x2_t {
    let c = _vdupq_n_u64(c);
    _veorq_u64(a, c)
}

#[inline(always)]
pub(crate) fn load_block<const RATE: usize>(s: &mut [[uint64x2_t; 5]; 5], blocks: [&[u8]; 2]) {
    debug_assert!(RATE <= blocks[0].len() && RATE % 8 == 0);
    for i in 0..RATE / 16 {
        let v0 = _vld1q_bytes_u64(&blocks[0][16 * i..16 * (i + 1)]);
        let v1 = _vld1q_bytes_u64(&blocks[1][16 * i..16 * (i + 1)]);
        s[(2 * i) / 5][(2 * i) % 5] = _veorq_u64(s[(2 * i) / 5][(2 * i) % 5], _vtrn1q_u64(v0, v1));
        s[(2 * i + 1) / 5][(2 * i + 1) % 5] =
            _veorq_u64(s[(2 * i + 1) / 5][(2 * i + 1) % 5], _vtrn2q_u64(v0, v1));
    }
    if RATE % 16 != 0 {
        let i = (RATE / 8 - 1) / 5;
        let j = (RATE / 8 - 1) % 5;
        let mut u = [0u64; 2];
        u[0] = u64::from_le_bytes(blocks[0][RATE - 8..RATE].try_into().unwrap());
        u[1] = u64::from_le_bytes(blocks[1][RATE - 8..RATE].try_into().unwrap());
        let uvec = _vld1q_u64(&u);
        s[i][j] = _veorq_u64(s[i][j], uvec);
    }
}

#[inline(always)]
pub(crate) fn load_block_full<const RATE: usize>(
    s: &mut [[uint64x2_t; 5]; 5],
    blocks: [[u8; 200]; 2],
) {
    load_block::<RATE>(s, [&blocks[0] as &[u8], &blocks[1] as &[u8]]);
}

#[inline(always)]
pub(crate) fn store_block<const RATE: usize>(s: &[[uint64x2_t; 5]; 5], out: [&mut [u8]; 2]) {
    for i in 0..RATE / 16 {
        let v0 = _vtrn1q_u64(
            s[(2 * i) / 5][(2 * i) % 5],
            s[(2 * i + 1) / 5][(2 * i + 1) % 5],
        );
        let v1 = _vtrn2q_u64(
            s[(2 * i) / 5][(2 * i) % 5],
            s[(2 * i + 1) / 5][(2 * i + 1) % 5],
        );
        _vst1q_bytes_u64(&mut out[0][16 * i..16 * (i + 1)], v0);
        _vst1q_bytes_u64(&mut out[1][16 * i..16 * (i + 1)], v1);
    }
    if RATE % 16 != 0 {
        debug_assert!(RATE % 8 == 0);
        let i = (RATE / 8 - 1) / 5;
        let j = (RATE / 8 - 1) % 5;
        let mut u = [0u8; 16];
        _vst1q_bytes_u64(&mut u, s[i][j]);
        out[0][RATE - 8..RATE].copy_from_slice(&u[0..8]);
        out[1][RATE - 8..RATE].copy_from_slice(&u[8..16]);
    }
}

#[inline(always)]
pub(crate) fn store_block_full<const RATE: usize>(s: &[[uint64x2_t; 5]; 5]) -> [[u8; 200]; 2] {
    let mut out0 = [0u8; 200];
    let mut out1 = [0u8; 200];
    store_block::<RATE>(s, [&mut out0, &mut out1]);
    [out0, out1]
}

#[inline(always)]
fn slice_2(a: [&[u8]; 2], start: usize, len: usize) -> [&[u8]; 2] {
    [&a[0][start..start + len], &a[1][start..start + len]]
}

#[inline(always)]
fn split_at_mut_2(out: [&mut [u8]; 2], mid: usize) -> ([&mut [u8]; 2], [&mut [u8]; 2]) {
    let [out0, out1] = out;
    let (out00, out01) = out0.split_at_mut(mid);
    let (out10, out11) = out1.split_at_mut(mid);
    ([out00, out10], [out01, out11])
}

impl KeccakItem<2> for uint64x2_t {
    #[inline(always)]
    fn zero() -> Self {
        _vdupq_n_u64(0)
    }
    #[inline(always)]
    fn xor5(a: Self, b: Self, c: Self, d: Self, e: Self) -> Self {
        _veor5q_u64(a, b, c, d, e)
    }
    #[inline(always)]
    fn rotate_left1_and_xor(a: Self, b: Self) -> Self {
        _vrax1q_u64(a, b)
    }
    #[inline(always)]
    fn xor_and_rotate<const LEFT: i32, const RIGHT: i32>(a: Self, b: Self) -> Self {
        _vxarq_u64::<LEFT, RIGHT>(a, b)
    }
    #[inline(always)]
    fn and_not_xor(a: Self, b: Self, c: Self) -> Self {
        _vbcaxq_u64(a, b, c)
    }
    #[inline(always)]
    fn xor_constant(a: Self, c: u64) -> Self {
        _veorq_n_u64(a, c)
    }
    #[inline(always)]
    fn xor(a: Self, b: Self) -> Self {
        _veorq_u64(a, b)
    }
    #[inline(always)]
    fn load_block<const BLOCKSIZE: usize>(a: &mut [[Self; 5]; 5], b: [&[u8]; 2]) {
        load_block::<BLOCKSIZE>(a, b)
    }
    #[inline(always)]
    fn store_block<const BLOCKSIZE: usize>(a: &[[Self; 5]; 5], b: [&mut [u8]; 2]) {
        store_block::<BLOCKSIZE>(a, b)
    }
    #[inline(always)]
    fn load_block_full<const BLOCKSIZE: usize>(a: &mut [[Self; 5]; 5], b: [[u8; 200]; 2]) {
        load_block_full::<BLOCKSIZE>(a, b)
    }
    #[inline(always)]
    fn store_block_full<const BLOCKSIZE: usize>(a: &[[Self; 5]; 5]) -> [[u8; 200]; 2] {
        store_block_full::<BLOCKSIZE>(a)
    }
    #[inline(always)]
    fn slice_n(a: [&[u8]; 2], start: usize, len: usize) -> [&[u8]; 2] {
        slice_2(a, start, len)
    }
    #[inline(always)]
    fn split_at_mut_n(a: [&mut [u8]; 2], mid: usize) -> ([&mut [u8]; 2], [&mut [u8]; 2]) {
        split_at_mut_2(a, mid)
    }
}