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
use super::element::UnsignedInteger;
pub struct MontgomeryAlgorithms;
impl MontgomeryAlgorithms {
/// Compute CIOS multiplication of `a` * `b`
/// `q` is the modulus
/// `mu` is the inverse of -q modulo 2^{64}
/// Notice CIOS stands for Coarsely Integrated Operand Scanning
/// For more information see section 2.3.2 of Tolga Acar's thesis
/// https://www.microsoft.com/en-us/research/wp-content/uploads/1998/06/97Acar.pdf
#[inline(always)]
pub const fn cios<const NUM_LIMBS: usize>(
a: &UnsignedInteger<NUM_LIMBS>,
b: &UnsignedInteger<NUM_LIMBS>,
q: &UnsignedInteger<NUM_LIMBS>,
mu: &u64,
) -> UnsignedInteger<NUM_LIMBS> {
let mut t = [0_u64; NUM_LIMBS];
let mut t_extra = [0_u64; 2];
let mut i: usize = NUM_LIMBS;
while i > 0 {
i -= 1;
// C := 0
let mut c: u128 = 0;
// for j=0 to N-1
// (C,t[j]) := t[j] + a[j]*b[i] + C
let mut cs: u128;
let mut j: usize = NUM_LIMBS;
while j > 0 {
j -= 1;
cs = t[j] as u128 + (a.limbs[j] as u128) * (b.limbs[i] as u128) + c;
c = cs >> 64;
t[j] = cs as u64;
}
// (t[N+1],t[N]) := t[N] + C
cs = (t_extra[1] as u128) + c;
t_extra[0] = (cs >> 64) as u64;
t_extra[1] = cs as u64;
let mut c: u128;
// m := t[0]*q'[0] mod D
let m = ((t[NUM_LIMBS - 1] as u128 * *mu as u128) << 64) >> 64;
// (C,_) := t[0] + m*q[0]
c = (t[NUM_LIMBS - 1] as u128 + m * (q.limbs[NUM_LIMBS - 1] as u128)) >> 64;
// for j=1 to N-1
// (C,t[j-1]) := t[j] + m*q[j] + C
let mut j: usize = NUM_LIMBS - 1;
while j > 0 {
j -= 1;
cs = t[j] as u128 + m * (q.limbs[j] as u128) + c;
c = cs >> 64;
t[j + 1] = ((cs << 64) >> 64) as u64;
}
// (C,t[N-1]) := t[N] + C
cs = (t_extra[1] as u128) + c;
c = cs >> 64;
t[0] = ((cs << 64) >> 64) as u64;
// t[N] := t[N+1] + C
t_extra[1] = t_extra[0] + c as u64;
}
let mut result = UnsignedInteger { limbs: t };
let overflow = t_extra[0] > 0;
if overflow || UnsignedInteger::const_le(q, &result) {
(result, _) = UnsignedInteger::sub(&result, q);
}
result
}
// Separated Operand Scanning Method (2.3.1)
#[inline(always)]
pub fn sos_square<const NUM_LIMBS: usize>(
a: &UnsignedInteger<NUM_LIMBS>,
q: &UnsignedInteger<NUM_LIMBS>,
mu: &u64,
) -> UnsignedInteger<NUM_LIMBS> {
// NOTE: we use explicit `while` loops in this function because profiling pointed
// at iterators of the form `(<x>..<y>).rev()` as the main performance bottleneck.
// Step 1: Compute `(hi, lo) = a * a`
let (mut hi, mut lo) = UnsignedInteger::square(a);
// Step 2: Add terms to `(hi, lo)` until multiple it
// is a multiple of both `2^{NUM_LIMBS * 64}` and
// `q`.
let mut c: u128 = 0;
let mut i = NUM_LIMBS;
while i > 0 {
i -= 1;
c = 0;
let m = (lo.limbs[i] as u128 * *mu as u128) as u64;
let mut j = NUM_LIMBS;
while j > 0 {
j -= 1;
if i + j >= NUM_LIMBS - 1 {
let index = i + j - (NUM_LIMBS - 1);
let cs = lo.limbs[index] as u128 + m as u128 * (q.limbs[j] as u128) + c;
c = cs >> 64;
lo.limbs[index] = cs as u64;
} else {
let index = i + j + 1;
let cs = hi.limbs[index] as u128 + m as u128 * (q.limbs[j] as u128) + c;
c = cs >> 64;
hi.limbs[index] = cs as u64;
}
}
// Carry propagation to `hi`
let mut t = 0;
while c > 0 && i >= t {
let cs = hi.limbs[i - t] as u128 + c;
c = cs >> 64;
hi.limbs[i - t] = cs as u64;
t += 1;
}
}
// Step 3: At this point `overflow * 2^{2 * NUM_LIMBS * 64} + (hi, lo)` is a multiple
// of `2^{NUM_LIMBS * 64}` and the result is obtained by dividing it by `2^{NUM_LIMBS * 64}`.
// In other words, `lo` is zero and the result is
// `overflow * 2^{NUM_LIMBS * 64} + hi`.
// That number is always strictly smaller than `2 * q`. To normalize it we substract
// `q` whenever it is larger than `q`.
// The easy case is when `overflow` is zero. We just use the `sub` function.
// If `overflow` is 1, then `hi` is smaller than `q`. The function `sub(hi, q)` wraps
// around `2^{NUM_LIMBS * 64}`. This is the result we need.
let overflow = c > 0;
if overflow || UnsignedInteger::const_le(q, &hi) {
(hi, _) = UnsignedInteger::sub(&hi, q);
}
hi
}
}
#[cfg(test)]
mod tests {
use crate::unsigned_integer::{element::U384, montgomery::MontgomeryAlgorithms};
#[test]
fn montgomery_multiplication_works_0() {
let x = U384::from_u64(11_u64);
let y = U384::from_u64(10_u64);
let m = U384::from_u64(23_u64); //
let mu: u64 = 3208129404123400281; // negative of the inverse of `m` modulo 2^{64}.
let c = U384::from_u64(13_u64); // x * y * (r^{-1}) % m, where r = 2^{64 * 6} and r^{-1} mod m = 2.
assert_eq!(MontgomeryAlgorithms::cios(&x, &y, &m, &mu), c);
}
#[test]
fn montgomery_multiplication_works_1() {
let x = U384::from_hex_unchecked("05ed176deb0e80b4deb7718cdaa075165f149c");
let y = U384::from_hex_unchecked("5f103b0bd4397d4df560eb559f38353f80eeb6");
let m = U384::from_hex_unchecked("cdb061954fdd36e5176f50dbdcfd349570a29ce1"); // this is prime
let mu: u64 = 16085280245840369887; // negative of the inverse of `m` modulo 2^{64}
let c = U384::from_hex_unchecked("8d65cdee621682815d59f465d2641eea8a1274dc"); // x * y * (r^{-1}) % m, where r = 2^{64 * 6}
assert_eq!(MontgomeryAlgorithms::cios(&x, &y, &m, &mu), c);
}
#[test]
fn montgomery_multiplication_works_3() {
let x = U384::from_hex_unchecked("8d65cdee621682815d59f465d2641eea8a1274dc");
let m = U384::from_hex_unchecked("cdb061954fdd36e5176f50dbdcfd349570a29ce1"); // this is prime
let r_mod_m = U384::from_hex_unchecked("58dfb0e1b3dd5e674bdcde4f42eb5533b8759d33");
let mu: u64 = 16085280245840369887; // negative of the inverse of `m` modulo 2^{64}
let c = U384::from_hex_unchecked("8d65cdee621682815d59f465d2641eea8a1274dc");
assert_eq!(MontgomeryAlgorithms::cios(&x, &r_mod_m, &m, &mu), c);
}
}