rind 0.1.3

Pure-Rust cryptography for MoQ, forked from graviola
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

// SPDX-License-Identifier: Apache-2.0 OR ISC OR MIT-0

//! Montgomery multiplication modulo an RSA modulus or prime.
//!
//! With BMI2 and ADX, 1024- and 2048-bit moduli run s2n-bignum's Karatsuba
//! routines and its 8n reduction. Otherwise, moduli of a multiple of four
//! words from eight run aws-lc's `bn_mul4x_mont`, and of a multiple of eight
//! its `bn_sqr8x_mont` for squaring, as aws-lc's `bn_mul_mont` picks them
//! without BMI2 and ADX; `admin/import-aws-lc` drops `bn_sqr8x_mont`'s branch
//! for those. They measured 1.5 to 1.9 times as fast as s2n-bignum's generic
//! `bignum_montmul` and `bignum_montsqr`, which run the other sizes.

use super::cpu::HaveBmi2Adx;

/// `z = x * y / 2^(64k) mod m`, for `x` and `y` below `m`, where `k` is the
/// length of each and `n0` is `-m^-1 mod 2^64`.
pub(crate) fn mont_mul(z: &mut [u64], x: &[u64], y: &[u64], m: &[u64], n0: u64) {
    debug_assert!(z.len() == m.len() && x.len() == m.len() && y.len() == m.len());
    match (HaveBmi2Adx::check(), m.len()) {
        (Some(_), 16 | 32) => karatsuba_mul(z, x, y, m, n0),
        (_, k) if k >= 8 && k.is_multiple_of(4) => mul4x(z, x, y, m, n0),
        _ => super::bignum_montmul::bignum_montmul(z, x, y, m),
    }
}

/// `z = x^2 / 2^(64k) mod m`, for `x` below `m`, where `k` is the length of
/// each and `n0` is `-m^-1 mod 2^64`.
pub(crate) fn mont_sqr(z: &mut [u64], x: &[u64], m: &[u64], n0: u64) {
    debug_assert!(z.len() == m.len() && x.len() == m.len());
    match (HaveBmi2Adx::check(), m.len()) {
        (Some(_), 16 | 32) => karatsuba_sqr(z, x, m, n0),
        (_, k) if k >= 8 && k.is_multiple_of(8) => sqr8x(z, x, m, n0),
        (_, k) if k >= 8 && k.is_multiple_of(4) => mul4x(z, x, x, m, n0),
        _ => super::bignum_montsqr::bignum_montsqr(z, x, m),
    }
}

/// `mont_mul` for 16 or 32 words, with BMI2 and ADX.
fn karatsuba_mul(z: &mut [u64], x: &[u64], y: &[u64], m: &[u64], n0: u64) {
    if m.len() == 16 {
        let mut product = [0; 32];
        super::bignum_kmul_16_32::bignum_kmul_16_32(&mut product, x, y, &mut [0; 32]);
        redc_8n(z, &mut product, m, n0);
    } else {
        let mut product = [0; 64];
        super::bignum_kmul_32_64::bignum_kmul_32_64(&mut product, x, y, &mut [0; 96]);
        redc_8n(z, &mut product, m, n0);
    }
}

/// `mont_sqr` for 16 or 32 words, with BMI2 and ADX.
fn karatsuba_sqr(z: &mut [u64], x: &[u64], m: &[u64], n0: u64) {
    if m.len() == 16 {
        let mut product = [0; 32];
        super::bignum_ksqr_16_32::bignum_ksqr_16_32(&mut product, x, &mut [0; 24]);
        redc_8n(z, &mut product, m, n0);
    } else {
        let mut product = [0; 64];
        super::bignum_ksqr_32_64::bignum_ksqr_32_64(&mut product, x, &mut [0; 72]);
        redc_8n(z, &mut product, m, n0);
    }
}

/// Reduces the `2k`-word `product` into the `k`-word `z`, fully, modulo `m`.
fn redc_8n(z: &mut [u64], product: &mut [u64], m: &[u64], n0: u64) {
    let carry = super::bignum_emontredc_8n::bignum_emontredc_8n(product, m, n0);
    let (_, reduced) = product.split_at(m.len());
    let carry = carry | super::bignum_cmp_lt::bignum_cmp_lt(m, reduced);
    super::bignum_optsub::bignum_optsub(z, reduced, m, carry);
}

/// `mont_mul` for a multiple of four words from eight.
fn mul4x(z: &mut [u64], x: &[u64], y: &[u64], m: &[u64], n0: u64) {
    let k = m.len();
    assert!(k >= 8 && k.is_multiple_of(4) && z.len() == k && x.len() == k && y.len() == k);
    // SAFETY: `z`, `x`, `y`, and `m` are each `k` words, as `bn_mul4x_mont`
    // reads and writes for such a `k`, and it reads `n0` as one word.
    unsafe { bn_mul4x_mont(z.as_mut_ptr(), x.as_ptr(), y.as_ptr(), m.as_ptr(), &n0, k) };
}

/// `mont_sqr` for a multiple of eight words.
fn sqr8x(z: &mut [u64], x: &[u64], m: &[u64], n0: u64) {
    let k = m.len();
    assert!(k >= 8 && k.is_multiple_of(8) && z.len() == k && x.len() == k);
    // SAFETY: `z`, `x`, and `m` are each `k` words, as `bn_sqr8x_mont` reads
    // and writes for such a `k`, and it reads `n0` as one word.
    unsafe { bn_sqr8x_mont(z.as_mut_ptr(), x.as_ptr(), 0, m.as_ptr(), &n0, k) };
}

// The routines take the System V ABI on every OS, since their text is
// aws-lc's ELF and Mach-O flavors, which read their arguments from its
// registers. Both return 1, which tells aws-lc's C they ran.

#[unsafe(naked)]
unsafe extern "sysv64" fn bn_mul4x_mont(
    rp: *mut u64,
    ap: *const u64,
    bp: *const u64,
    np: *const u64,
    n0: *const u64,
    num: usize,
) -> i32 {
    #[cfg(not(target_vendor = "apple"))]
    core::arch::naked_asm!(
        include_str!("aws_lc/bn_mul4x_mont.linux.s"),
        options(att_syntax)
    );
    #[cfg(target_vendor = "apple")]
    core::arch::naked_asm!(
        include_str!("aws_lc/bn_mul4x_mont.apple.s"),
        options(att_syntax)
    );
}

/// `mulx_adx_capable` selects nothing: the import drops that branch.
#[unsafe(naked)]
unsafe extern "sysv64" fn bn_sqr8x_mont(
    rp: *mut u64,
    ap: *const u64,
    mulx_adx_capable: u64,
    np: *const u64,
    n0: *const u64,
    num: usize,
) -> i32 {
    #[cfg(not(target_vendor = "apple"))]
    core::arch::naked_asm!(
        include_str!("aws_lc/bn_sqr8x_mont.linux.s"),
        bn_sqr8x_internal = sym bn_sqr8x_internal,
        options(att_syntax)
    );
    #[cfg(target_vendor = "apple")]
    core::arch::naked_asm!(
        include_str!("aws_lc/bn_sqr8x_mont.apple.s"),
        bn_sqr8x_internal = sym bn_sqr8x_internal,
        options(att_syntax)
    );
}

/// `bn_sqr8x_mont`'s squaring and reduction, which takes its arguments in
/// registers and a stack frame of that routine's making, so only it calls
/// this.
#[unsafe(naked)]
unsafe extern "sysv64" fn bn_sqr8x_internal() {
    #[cfg(not(target_vendor = "apple"))]
    core::arch::naked_asm!(
        include_str!("aws_lc/bn_sqr8x_internal.linux.s"),
        options(att_syntax)
    );
    #[cfg(target_vendor = "apple")]
    core::arch::naked_asm!(
        include_str!("aws_lc/bn_sqr8x_internal.apple.s"),
        options(att_syntax)
    );
}

#[cfg(test)]
#[cfg_attr(coverage_nightly, coverage(off))]
mod tests {
    use super::super::adx_mux::tests::Rng;
    use super::super::{bignum_montmul, bignum_montsqr, bignum_negmodinv};
    use super::*;

    // Every other test runs only the routines this CPU picks, so here aws-lc's
    // are checked against s2n-bignum's generic ones at each size they take,
    // and so are the Karatsuba routines where the CPU has BMI2 and ADX.
    #[test]
    fn matches_generic() {
        let adx = HaveBmi2Adx::check().is_some();
        let mut rng = Rng(0x2545_f491_4f6c_dd1d);
        for k in [8, 12, 16, 24, 32, 48, 64] {
            for case in 0..16 {
                let mut m = rng.words::<64>(64 * k as u32);
                let m = &mut m[..k];
                // The widest modulus, the narrowest with its top word set,
                // and random ones.
                match case {
                    0 => m.fill(u64::MAX),
                    1 => {
                        m.fill(0);
                        m[k - 1] = 1;
                    }
                    _ => m[k - 1] |= 1 << 63,
                }
                m[0] |= 1;
                let mut n0 = [0];
                bignum_negmodinv::bignum_negmodinv(&mut n0, &m[..1]);
                let n0 = n0[0];

                // Operands below `m`, with its extremes: zero, and `m - 1`
                // for both.
                let below = |rng: &mut Rng| {
                    let mut x = rng.words::<64>(64 * k as u32);
                    x[k - 1] %= m[k - 1];
                    x
                };
                let (mut x, mut y) = (below(&mut rng), below(&mut rng));
                match case {
                    2 => x[..k].fill(0),
                    3 => {
                        x[..k].copy_from_slice(m);
                        x[0] -= 1;
                        y = x;
                    }
                    _ => {}
                }
                let (x, y) = (&x[..k], &y[..k]);

                let mut want = [0; 64];
                let want = &mut want[..k];
                let mut got = [0; 64];
                let got = &mut got[..k];

                bignum_montmul::bignum_montmul(want, x, y, m);
                mul4x(got, x, y, m, n0);
                assert_eq!(want, got, "mul4x, {k} words, case {case}");
                if adx && (k == 16 || k == 32) {
                    karatsuba_mul(got, x, y, m, n0);
                    assert_eq!(want, got, "karatsuba_mul, {k} words, case {case}");
                }

                bignum_montsqr::bignum_montsqr(want, x, m);
                mul4x(got, x, x, m, n0);
                assert_eq!(want, got, "mul4x squaring, {k} words, case {case}");
                if k.is_multiple_of(8) {
                    sqr8x(got, x, m, n0);
                    assert_eq!(want, got, "sqr8x, {k} words, case {case}");
                }
                if adx && (k == 16 || k == 32) {
                    karatsuba_sqr(got, x, m, n0);
                    assert_eq!(want, got, "karatsuba_sqr, {k} words, case {case}");
                }
            }
        }
    }
}