arnaucube/poulpy
1
0
Fork 0
mirror of https://github.com/arnaucube/poulpy.git synced 2026-02-10 05:06:44 +01:00
Code Issues Packages Projects Releases Wiki Activity

Added const for modular reduction, simplfied API

Browse Source
This commit is contained in:
Jean-Philippe Bossuat
2024-12-30 16:20:25 +01:00
parent 7344d78173
commit c65522099b
9 changed files with 227 additions and 162 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
benches/operations.rs
View File

@@ -1,6 +1,10 @@
use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion};
use math::ring::Ring;
use math::modulus::Operations;
use math::modulus::montgomery::Montgomery;
use math::modulus::{NONE, ONCE};
const CHUNK: usize= 8;
fn add_vec_unary(c: &mut Criterion) {
fn runner(r: Ring<u64>) -> Box<dyn FnMut()> {
@@ -11,9 +15,8 @@ fn add_vec_unary(c: &mut Criterion) {
p0.0[i] = i as u64;
p1.0[i] = i as u64;
}
println!("{}", r.n());
Box::new(move || {
r.modulus.add_vec_unary_assign::<8>(&p0.0, &mut p1.0);
r.modulus.add_vec_unary_assign::<CHUNK, ONCE>(&p0.0, &mut p1.0);
})
}
@@ -36,5 +39,38 @@ fn add_vec_unary(c: &mut Criterion) {
}
}
criterion_group!(benches, add_vec_unary);
fn mul_vec_montgomery_external_unary_assign(c: &mut Criterion) {
fn runner(r: Ring<u64>) -> Box<dyn FnMut()> {
let mut p0: math::poly::Poly<Montgomery<u64>> = r.new_poly_montgomery();
let mut p1: math::poly::Poly<u64> = r.new_poly();
for i in 0..p0.n(){
p0.0[i] = r.modulus.montgomery.prepare::<ONCE>(i as u64);
p1.0[i] = i as u64;
}
Box::new(move || {
r.modulus.mul_vec_montgomery_external_unary_assign::<CHUNK, NONE>(&p0.0, &mut p1.0);
})
}
let mut b: criterion::BenchmarkGroup<'_, criterion::measurement::WallTime> = c.benchmark_group("mul_vec_montgomery_external_unary_assign");
for log_n in 11..17 {
let n: usize = 1<<log_n as usize;
let q_base: u64 = 0x1fffffffffe00001u64;
let q_power: usize = 1usize;
let r: Ring<u64> = Ring::<u64>::new(n, q_base, q_power);
let runners = [
("prime", {
runner(r)
}),
];
for (name, mut runner) in runners {
let id = BenchmarkId::new(name, n);
b.bench_with_input(id, &(), |b, _| b.iter(&mut runner));
}
}
}
criterion_group!(benches, add_vec_unary, mul_vec_montgomery_external_unary_assign);
criterion_main!(benches);

9
src/dft/ntt.rs
View File

@@ -3,6 +3,7 @@ use crate::modulus::shoup::Shoup;
use crate::modulus::prime::Prime;
use crate::modulus::ReduceOnce;
use crate::modulus::WordOps;
use crate::modulus::ONCE;
use crate::dft::DFT;
use itertools::izip;
@@ -22,7 +23,7 @@ impl Table< u64> {
let psi: u64 = prime.primitive_nth_root(nth_root);
let psi_mont: Montgomery<u64> = prime.montgomery.prepare(psi);
let psi_mont: Montgomery<u64> = prime.montgomery.prepare::<ONCE>(psi);
let psi_inv_mont: Montgomery<u64> = prime.montgomery.pow(psi_mont, prime.phi-1);
let mut psi_forward_rev: Vec<Shoup<u64>> = vec![Shoup(0, 0); (nth_root >> 1) as usize];
@@ -40,8 +41,8 @@ impl Table< u64> {
let i_rev: usize = i.reverse_bits_msb(log_nth_root_half);
prime.montgomery.mul_external_assign(psi_mont, &mut powers_forward);
prime.montgomery.mul_external_assign(psi_inv_mont, &mut powers_backward);
prime.montgomery.mul_external_assign::<ONCE>(psi_mont, &mut powers_forward);
prime.montgomery.mul_external_assign::<ONCE>(psi_inv_mont, &mut powers_backward);
psi_forward_rev[i_rev] = prime.shoup.prepare(powers_forward);
psi_backward_rev[i_rev] = prime.shoup.prepare(powers_backward);
@@ -61,7 +62,7 @@ impl Table< u64> {
// Returns n^-1 mod q in Montgomery.
fn inv(&self, n:u64) -> Montgomery<u64>{
self.prime.montgomery.pow(self.prime.montgomery.prepare(n), self.prime.phi-1)
self.prime.montgomery.pow(self.prime.montgomery.prepare::<ONCE>(n), self.prime.phi-1)
}
}

45
src/modulus.rs
View File

@@ -4,7 +4,14 @@ pub mod montgomery;
pub mod shoup;
pub mod impl_u64;
pub type REDUCEMOD = u8;
pub const NONE: REDUCEMOD = 0;
pub const ONCE: REDUCEMOD = 1;
pub const TWICE: REDUCEMOD = 2;
pub const FOURTIMES: REDUCEMOD = 3;
pub const BARRETT: REDUCEMOD = 4;
pub const BARRETTLAZY: REDUCEMOD = 5;
pub trait WordOps<O>{
fn log2(self) -> O;
@@ -64,47 +71,61 @@ impl ReduceOnce<u64> for u64{
#[inline(always)]
fn reduce_once_assign(&mut self, q: u64){
debug_assert!(q < 0x8000000000000000, "2q >= 2^64");
*self = (*self).min(self.wrapping_sub(q))
*self = *self.min(&mut self.wrapping_sub(q))
}
#[inline(always)]
fn reduce_once(&self, q:u64) -> u64{
debug_assert!(q < 0x8000000000000000, "2q >= 2^64");
(*self).min(self.wrapping_sub(q))
*self.min(&mut self.wrapping_sub(q))
}
}
pub trait Operations<O>{
// Assigns a + b to c.
fn add_binary_assign(&self, a: &O, b:&O, c: &mut O);
fn add_binary_assign<const REDUCE:REDUCEMOD>(&self, a: &O, b:&O, c: &mut O);
// Assigns a + b to b.
fn add_unary_assign(&self, a: &O, b: &mut O);
fn add_unary_assign<const REDUCE:REDUCEMOD>(&self, a: &O, b: &mut O);
// Assigns a[i] + b[i] to c[i]
fn add_vec_binary_assign<const CHUNK:usize>(&self, a: &[O], b:&[O], c: &mut [O]);
fn add_vec_binary_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a: &[O], b:&[O], c: &mut [O]);
// Assigns a[i] + b[i] to b[i]
fn add_vec_unary_assign<const CHUNK:usize>(&self, a: &[O], b: &mut [O]);
fn add_vec_unary_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a: &[O], b: &mut [O]);
// Assigns a - b to c.
fn sub_binary_assign(&self, a: &O, b:&O, c: &mut O);
fn sub_binary_assign<const REDUCE:REDUCEMOD>(&self, a: &O, b:&O, c: &mut O);
// Assigns b - a to b.
fn sub_unary_assign(&self, a: &O, b: &mut O);
fn sub_unary_assign<const REDUCE:REDUCEMOD>(&self, a: &O, b: &mut O);
// Assigns a[i] - b[i] to c[i]
fn sub_vec_binary_assign<const CHUNK:usize>(&self, a: &[O], b:&[O], c: &mut [O]);
fn sub_vec_binary_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a: &[O], b:&[O], c: &mut [O]);
// Assigns a[i] - b[i] to b[i]
fn sub_vec_unary_assign<const CHUNK:usize>(&self, a: &[O], b: &mut [O]);
fn sub_vec_unary_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a: &[O], b: &mut [O]);
// Assigns -a to a.
fn neg_assign(&self, a:&mut O);
fn neg_assign<const REDUCE:REDUCEMOD>(&self, a:&mut O);
// Assigns -a[i] to a[i].
fn neg_vec_assign<const CHUNK:usize>(&self, a: &mut [O]);
fn neg_vec_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a: &mut [O]);
// Assigns a * b to c.
fn mul_montgomery_external_binary_assign<const REDUCE:REDUCEMOD>(&self, a:&montgomery::Montgomery<O>, b:&O, c: &mut O);
// Assigns a * b to b.
fn mul_montgomery_external_unary_assign<const REDUCE:REDUCEMOD>(&self, a:&montgomery::Montgomery<O>, b:&mut O);
// Assigns a[i] * b[i] to c[i].
fn mul_vec_montgomery_external_binary_assign<const CHUNK:usize,const REDUCE:REDUCEMOD>(&self, a:&[montgomery::Montgomery<O>], b:&[O], c: &mut [O]);
// Assigns a[i] * b[i] to b[i].
fn mul_vec_montgomery_external_unary_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a:&[montgomery::Montgomery<O>], b:&mut [O]);
}

164
src/modulus/impl_u64/montgomery.rs
View File

@@ -1,7 +1,8 @@
use crate::modulus::ReduceOnce;
use crate::modulus::montgomery::{MontgomeryPrecomp, Montgomery};
use crate::modulus::barrett::{BarrettPrecomp};
use crate::modulus::barrett::BarrettPrecomp;
use crate::modulus::{REDUCEMOD, NONE, ONCE, TWICE, FOURTIMES, BARRETT, BARRETTLAZY};
extern crate test;
/// MontgomeryPrecomp is a set of methods implemented for MontgomeryPrecomp<u64>
@@ -21,13 +22,15 @@ impl MontgomeryPrecomp<u64>{
}
let mut precomp = Self{
q: q,
two_q: q<<1,
four_q: q<<2,
barrett: BarrettPrecomp::new(q),
q_inv: q_inv,
one: Montgomery(0),
minus_one: Montgomery(0),
};
precomp.one = precomp.prepare(1);
precomp.one = precomp.prepare::<ONCE>(1);
precomp.minus_one = Montgomery(q-precomp.one.value());
precomp
@@ -45,120 +48,91 @@ impl MontgomeryPrecomp<u64>{
self.minus_one
}
/// Applies a modular reduction on x based on REDUCE:
/// - LAZY: no modular reduction.
/// - ONCE: subtracts q if x >= q.
/// - FULL: maps x to x mod q using Barrett reduction.
#[inline(always)]
pub fn reduce<const REDUCE:REDUCEMOD>(&self, x: u64) -> u64{
let mut r: u64 = x;
self.reduce_assign::<REDUCE>(&mut r);
r
}
/// Applies a modular reduction on x based on REDUCE:
/// - LAZY: no modular reduction.
/// - ONCE: subtracts q if x >= q.
/// - FULL: maps x to x mod q using Barrett reduction.
#[inline(always)]
pub fn reduce_assign<const REDUCE:REDUCEMOD>(&self, x: &mut u64){
match REDUCE {
NONE =>{},
ONCE =>{x.reduce_once_assign(self.q)},
TWICE=>{x.reduce_once_assign(self.two_q)},
FOURTIMES =>{x.reduce_once_assign(self.four_q)},
BARRETT =>{self.barrett.reduce_assign(x)},
BARRETTLAZY =>{self.barrett.reduce_assign(x)},
_ => unreachable!("invalid REDUCE argument")
}
}
/// Returns lhs * 2^64 mod q as a Montgomery<u64>.
#[inline(always)]
pub fn prepare(&self, lhs: u64) -> Montgomery<u64>{
pub fn prepare<const REDUCE:REDUCEMOD>(&self, lhs: u64) -> Montgomery<u64>{
let mut rhs = Montgomery(0);
self.prepare_assign(lhs, &mut rhs);
self.prepare_assign::<REDUCE>(lhs, &mut rhs);
rhs
}
/// Assigns lhs * 2^64 mod q to rhs.
#[inline(always)]
pub fn prepare_assign(&self, lhs: u64, rhs: &mut Montgomery<u64>){
self.prepare_lazy_assign(lhs, rhs);
rhs.value_mut().reduce_once_assign(self.q);
}
/// Returns lhs * 2^64 mod q in range [0, 2q-1] as a Montgomery<u64>.
#[inline(always)]
pub fn prepare_lazy(&self, lhs: u64) -> Montgomery<u64>{
let mut rhs = Montgomery(0);
self.prepare_lazy_assign(lhs, &mut rhs);
rhs
}
/// Assigns lhs * 2^64 mod q in range [0, 2q-1] to rhs.
#[inline(always)]
pub fn prepare_lazy_assign(&self, lhs: u64, rhs: &mut Montgomery<u64>){
pub fn prepare_assign<const REDUCE:REDUCEMOD>(&self, lhs: u64, rhs: &mut Montgomery<u64>){
let (_, mhi) = lhs.widening_mul(*self.barrett.value_lo());
*rhs = Montgomery((lhs.wrapping_mul(*self.barrett.value_hi()).wrapping_add(mhi)).wrapping_mul(self.q).wrapping_neg());
self.reduce_assign::<REDUCE>(rhs.value_mut());
}
/// Returns lhs * (2^64)^-1 mod q as a u64.
#[inline(always)]
pub fn unprepare(&self, lhs: Montgomery<u64>) -> u64{
pub fn unprepare<const REDUCE:REDUCEMOD>(&self, lhs: Montgomery<u64>) -> u64{
let mut rhs = 0u64;
self.unprepare_assign(lhs, &mut rhs);
self.unprepare_assign::<REDUCE>(lhs, &mut rhs);
rhs
}
/// Assigns lhs * (2^64)^-1 mod q to rhs.
#[inline(always)]
pub fn unprepare_assign(&self, lhs: Montgomery<u64>, rhs: &mut u64){
self.unprepare_lazy_assign(lhs, rhs);
rhs.reduce_once_assign(self.q);
}
/// Returns lhs * (2^64)^-1 mod q in range [0, 2q-1].
#[inline(always)]
pub fn unprepare_lazy(&self, lhs: Montgomery<u64>) -> u64{
let mut rhs = 0u64;
self.unprepare_lazy_assign(lhs, &mut rhs);
rhs
}
/// Assigns lhs * (2^64)^-1 mod q in range [0, 2q-1] to rhs.
#[inline(always)]
pub fn unprepare_lazy_assign(&self, lhs: Montgomery<u64>, rhs: &mut u64){
pub fn unprepare_assign<const REDUCE:REDUCEMOD>(&self, lhs: Montgomery<u64>, rhs: &mut u64){
let (_, r) = self.q.widening_mul(lhs.value().wrapping_mul(self.q_inv));
*rhs = self.q - r
*rhs = self.reduce::<REDUCE>(self.q.wrapping_sub(r));
}
/// Returns lhs * rhs * (2^{64})^-1 mod q.
#[inline(always)]
pub fn mul_external(&self, lhs: Montgomery<u64>, rhs: u64) -> u64{
let mut r = self.mul_external_lazy(lhs, rhs);
r.reduce_once_assign(self.q);
pub fn mul_external<const REDUCE:REDUCEMOD>(&self, lhs: Montgomery<u64>, rhs: u64) -> u64{
let mut r: u64 = rhs;
self.mul_external_assign::<REDUCE>(lhs, &mut r);
r
}
/// Assigns lhs * rhs * (2^{64})^-1 mod q to rhs.
#[inline(always)]
pub fn mul_external_assign(&self, lhs: Montgomery<u64>, rhs: &mut u64){
self.mul_external_lazy_assign(lhs, rhs);
rhs.reduce_once_assign(self.q);
}
/// Returns lhs * rhs * (2^{64})^-1 mod q in range [0, 2q-1].
#[inline(always)]
pub fn mul_external_lazy(&self, lhs: Montgomery<u64>, rhs: u64) -> u64{
let mut result: u64 = rhs;
self.mul_external_lazy_assign(lhs, &mut result);
result
}
/// Assigns lhs * rhs * (2^{64})^-1 mod q in range [0, 2q-1] to rhs.
#[inline(always)]
pub fn mul_external_lazy_assign(&self, lhs: Montgomery<u64>, rhs: &mut u64){
pub fn mul_external_assign<const REDUCE:REDUCEMOD>(&self, lhs: Montgomery<u64>, rhs: &mut u64){
let (mlo, mhi) = lhs.value().widening_mul(*rhs);
let (_, hhi) = self.q.widening_mul(mlo.wrapping_mul(self.q_inv));
*rhs = mhi.wrapping_add(self.q - hhi)
*rhs = self.reduce::<REDUCE>(mhi.wrapping_sub(hhi).wrapping_add(self.q));
}
/// Returns lhs * rhs * (2^{64})^-1 mod q in range [0, 2q-1].
#[inline(always)]
pub fn mul_internal(&self, lhs: Montgomery<u64>, rhs: Montgomery<u64>) -> Montgomery<u64>{
Montgomery(self.mul_external(lhs, *rhs.value()))
pub fn mul_internal<const REDUCE:REDUCEMOD>(&self, lhs: Montgomery<u64>, rhs: Montgomery<u64>) -> Montgomery<u64>{
Montgomery(self.mul_external::<REDUCE>(lhs, *rhs.value()))
}
/// Assigns lhs * rhs * (2^{64})^-1 mod q to rhs.
#[inline(always)]
pub fn mul_internal_assign(&self, lhs: Montgomery<u64>, rhs: &mut Montgomery<u64>){
self.mul_external_assign(lhs, rhs.value_mut());
}
/// Returns lhs * rhs * (2^{64})^-1 mod q in range [0, 2q-1].
#[inline(always)]
pub fn mul_internal_lazy(&self, lhs: Montgomery<u64>, rhs: Montgomery<u64>) -> Montgomery<u64>{
Montgomery(self.mul_external_lazy(lhs, *rhs.value()))
}
/// Assigns lhs * rhs * (2^{64})^-1 mod q in range [0, 2q-1] to rhs.
#[inline(always)]
pub fn mul_internal_lazy_assign(&self, lhs: Montgomery<u64>, rhs: &mut Montgomery<u64>){
self.mul_external_lazy_assign(lhs, rhs.value_mut());
pub fn mul_internal_assign<const REDUCE:REDUCEMOD>(&self, lhs: Montgomery<u64>, rhs: &mut Montgomery<u64>){
self.mul_external_assign::<REDUCE>(lhs, rhs.value_mut());
}
#[inline(always)]
@@ -174,27 +148,11 @@ impl MontgomeryPrecomp<u64>{
/// Assigns lhs + rhs - q if (lhs + rhs) >= q to rhs.
#[inline(always)]
pub fn add_internal_reduce_once_assign(&self, lhs: Montgomery<u64>, rhs: &mut Montgomery<u64>){
pub fn add_internal_reduce_once_assign<const LAZY:bool>(&self, lhs: Montgomery<u64>, rhs: &mut Montgomery<u64>){
self.add_internal_lazy_assign(lhs, rhs);
rhs.value_mut().reduce_once_assign(self.q);
}
#[inline(always)]
pub fn reduce(&self, lhs: u64) -> u64{
self.barrett.reduce(lhs)
}
/// Returns lhs mod q in range [0, 2q-1].
#[inline(always)]
pub fn reduce_lazy(&self, lhs: u64) -> u64{
self.barrett.reduce_lazy(lhs)
}
#[inline(always)]
pub fn reduce_assign(&self, lhs: &mut u64){
self.barrett.reduce_assign(lhs)
}
/// Returns lhs mod q in range [0, 2q-1].
#[inline(always)]
pub fn reduce_lazy_assign(&self, lhs: &mut u64){
@@ -209,9 +167,9 @@ impl MontgomeryPrecomp<u64>{
let mut i: u64 = exponent;
while i > 0{
if i & 1 == 1{
self.mul_internal_assign(x_mut, &mut y);
self.mul_internal_assign::<ONCE>(x_mut, &mut y);
}
self.mul_internal_assign(x_mut, &mut x_mut);
self.mul_internal_assign::<ONCE>(x_mut, &mut x_mut);
i >>= 1;
}
@@ -227,16 +185,16 @@ impl MontgomeryPrecomp<u64>{
fn pow(x:u64, exponent:u64, q:u64) -> u64{
let montgomery: MontgomeryPrecomp<u64> = MontgomeryPrecomp::<u64>::new(q);
let mut y_mont: Montgomery<u64> = montgomery.one();
let mut x_mont: Montgomery<u64> = montgomery.prepare(x);
let mut x_mont: Montgomery<u64> = montgomery.prepare::<ONCE>(x);
while exponent > 0{
if exponent & 1 == 1{
montgomery.mul_internal_assign(x_mont, &mut y_mont);
montgomery.mul_internal_assign::<ONCE>(x_mont, &mut y_mont);
}
montgomery.mul_internal_assign(x_mont, &mut x_mont);
montgomery.mul_internal_assign::<ONCE>(x_mont, &mut x_mont);
}
montgomery.unprepare(y_mont)
montgomery.unprepare::<ONCE>(y_mont)
}
#[cfg(test)]
@@ -251,8 +209,8 @@ mod tests {
let m_precomp = montgomery::MontgomeryPrecomp::new(q);
let x: u64 = 0x5f876e514845cc8b;
let y: u64 = 0xad726f98f24a761a;
let y_mont = m_precomp.prepare(y);
assert!(m_precomp.mul_external(y_mont, x) == (x as u128 * y as u128 % q as u128) as u64);
let y_mont = m_precomp.prepare::<ONCE>(y);
assert!(m_precomp.mul_external::<ONCE>(y_mont, x) == (x as u128 * y as u128 % q as u128) as u64);
}
#[bench]
@@ -261,7 +219,7 @@ mod tests {
let m_precomp = montgomery::MontgomeryPrecomp::new(q);
let mut x: u64 = 0x5f876e514845cc8b;
let y: u64 = 0xad726f98f24a761a;
let y_mont = m_precomp.prepare(y);
b.iter(|| m_precomp.mul_external_assign(y_mont, &mut x));
let y_mont = m_precomp.prepare::<ONCE>(y);
b.iter(|| m_precomp.mul_external_assign::<ONCE>(y_mont, &mut x));
}
}

60
src/modulus/impl_u64/operations.rs
View File

@@ -2,59 +2,83 @@
use crate::modulus::Operations;
use crate::modulus::prime::Prime;
use crate::modulus::ReduceOnce;
use crate::modulus::montgomery::Montgomery;
use crate::modulus::{REDUCEMOD, NONE, ONCE, BARRETT, BARRETTLAZY};
use crate::{apply_unary, apply_binary, apply_ternary};
use itertools::izip;
impl Operations<u64> for Prime<u64>{
#[inline(always)]
fn add_binary_assign(&self, a: &u64, b: &u64, c: &mut u64){
*c = a.wrapping_add(*b).reduce_once(self.q);
fn add_binary_assign<const REDUCE:REDUCEMOD>(&self, a: &u64, b: &u64, c: &mut u64){
*c = a.wrapping_add(*b);
self.montgomery.reduce_assign::<REDUCE>(c);
}
#[inline(always)]
fn add_unary_assign(&self, a: &u64, b: &mut u64){
*b = a.wrapping_add(*b).reduce_once(self.q);
fn add_unary_assign<const REDUCE:REDUCEMOD>(&self, a: &u64, b: &mut u64){
*b = a.wrapping_add(*b);
self.montgomery.reduce_assign::<REDUCE>(b);
}
#[inline(always)]
fn add_vec_binary_assign<const CHUNK:usize>(&self, a: &[u64], b:&[u64], c:&mut [u64]){
apply_ternary!(self, Self::add_binary_assign, a, b, c, CHUNK);
fn add_vec_binary_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a: &[u64], b:&[u64], c:&mut [u64]){
apply_ternary!(self, Self::add_binary_assign::<REDUCE>, a, b, c, CHUNK);
}
#[inline(always)]
fn add_vec_unary_assign<const CHUNK:usize>(&self, a: &[u64], b:&mut [u64]){
apply_binary!(self, Self::add_unary_assign, a, b, CHUNK);
fn add_vec_unary_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a: &[u64], b:&mut [u64]){
apply_binary!(self, Self::add_unary_assign::<REDUCE>, a, b, CHUNK);
}
#[inline(always)]
fn sub_binary_assign(&self, a: &u64, b: &u64, c: &mut u64){
fn sub_binary_assign<const REDUCE:REDUCEMOD>(&self, a: &u64, b: &u64, c: &mut u64){
*c = a.wrapping_add(self.q.wrapping_sub(*b)).reduce_once(self.q);
}
#[inline(always)]
fn sub_unary_assign(&self, a: &u64, b: &mut u64){
fn sub_unary_assign<const REDUCE:REDUCEMOD>(&self, a: &u64, b: &mut u64){
*b = a.wrapping_add(self.q.wrapping_sub(*b)).reduce_once(self.q);
}
#[inline(always)]
fn sub_vec_binary_assign<const CHUNK:usize>(&self, a: &[u64], b:&[u64], c:&mut [u64]){
apply_ternary!(self, Self::sub_binary_assign, a, b, c, CHUNK);
fn sub_vec_binary_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a: &[u64], b:&[u64], c:&mut [u64]){
apply_ternary!(self, Self::sub_binary_assign::<REDUCE>, a, b, c, CHUNK);
}
#[inline(always)]
fn sub_vec_unary_assign<const CHUNK:usize>(&self, a: &[u64], b:&mut [u64]){
apply_binary!(self, Self::sub_unary_assign, a, b, CHUNK);
fn sub_vec_unary_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a: &[u64], b:&mut [u64]){
apply_binary!(self, Self::sub_unary_assign::<REDUCE>, a, b, CHUNK);
}
#[inline(always)]
fn neg_assign(&self, a: &mut u64){
fn neg_assign<const REDUCE:REDUCEMOD>(&self, a: &mut u64){
*a = self.q.wrapping_sub(*a);
self.montgomery.reduce_assign::<REDUCE>(a)
}
#[inline(always)]
fn neg_vec_assign<const CHUNK:usize>(&self, a: &mut [u64]){
apply_unary!(self, Self::neg_assign, a, CHUNK);
}
fn neg_vec_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a: &mut [u64]){
apply_unary!(self, Self::neg_assign::<REDUCE>, a, CHUNK);
}
#[inline(always)]
fn mul_montgomery_external_binary_assign<const REDUCE:REDUCEMOD>(&self, a:& Montgomery<u64>, b:&u64, c: &mut u64){
*c = self.montgomery.mul_external::<REDUCE>(*a, *b);
}
#[inline(always)]
fn mul_montgomery_external_unary_assign<const REDUCE:REDUCEMOD>(&self, lhs:&Montgomery<u64>, rhs:&mut u64){
*rhs = self.montgomery.mul_external::<REDUCE>(*lhs, *rhs);
}
#[inline(always)]
fn mul_vec_montgomery_external_binary_assign<const CHUNK:usize,const REDUCE:REDUCEMOD>(&self, a:& [Montgomery<u64>], b:&[u64], c: &mut [u64]){
apply_ternary!(self, Self::mul_montgomery_external_binary_assign::<REDUCE>, a, b, c, CHUNK);
}
#[inline(always)]
fn mul_vec_montgomery_external_unary_assign<const CHUNK:usize, const REDUCE:REDUCEMOD>(&self, a:&[Montgomery<u64>], b:&mut [u64]){
apply_binary!(self, Self::mul_montgomery_external_unary_assign::<REDUCE>, a, b, CHUNK);
}
}

31
src/modulus/impl_u64/prime.rs
View File

@@ -1,6 +1,7 @@
use crate::modulus::prime::Prime;
use crate::modulus::montgomery::{Montgomery, MontgomeryPrecomp};
use crate::modulus::shoup::{ShoupPrecomp};
use crate::modulus::ONCE;
use primality_test::is_prime;
use prime_factorization::Factorization;
@@ -63,20 +64,20 @@ impl Prime<u64>{
#[inline(always)]
pub fn pow(&self, x: u64, exponent: u64) -> u64{
let mut y_mont: Montgomery<u64> = self.montgomery.one();
let mut x_mont: Montgomery<u64> = self.montgomery.prepare(x);
let mut x_mont: Montgomery<u64> = self.montgomery.prepare::<ONCE>(x);
let mut i: u64 = exponent;
while i > 0{
if i & 1 == 1{
self.montgomery.mul_internal_assign(x_mont, &mut y_mont);
self.montgomery.mul_internal_assign::<ONCE>(x_mont, &mut y_mont);
}
self.montgomery.mul_internal_assign(x_mont, &mut x_mont);
self.montgomery.mul_internal_assign::<ONCE>(x_mont, &mut x_mont);
i >>= 1;
}
self.montgomery.unprepare(y_mont)
self.montgomery.unprepare::<ONCE>(y_mont)
}
/// Returns x^-1 mod q.
@@ -165,8 +166,6 @@ impl Prime<u64>{
panic!("invalid factor list: does not fully divide q_base: q_base % (all factors) = {}", q_base)
}
}
}
/// Returns (psi + a * q_base)^{nth_root} = 1 mod q = q_base^q_power given psi^{nth_root} = 1 mod q_base.
@@ -174,23 +173,23 @@ impl Prime<u64>{
fn hensel_lift(&self, psi: u64, nth_root: u64) -> u64{
assert!(Pow(psi, nth_root, self.q_base)==1, "invalid argument psi: psi^nth_root = {} != 1", Pow(psi, nth_root, self.q_base));
let mut psi_mont: Montgomery<u64> = self.montgomery.prepare(psi);
let nth_root_mont: Montgomery<u64> = self.montgomery.prepare(nth_root);
let mut psi_mont: Montgomery<u64> = self.montgomery.prepare::<ONCE>(psi);
let nth_root_mont: Montgomery<u64> = self.montgomery.prepare::<ONCE>(nth_root);
for _i in 1..self.q_power{
let psi_pow: Montgomery<u64> = self.montgomery.pow(psi_mont, nth_root-1);
let num: Montgomery<u64> = Montgomery(self.montgomery.one().value() + self.q - self.montgomery.mul_internal(psi_pow, psi_mont).value());
let num: Montgomery<u64> = Montgomery(self.montgomery.one().value() + self.q - self.montgomery.mul_internal::<ONCE>(psi_pow, psi_mont).value());
let mut den: Montgomery<u64> = self.montgomery.mul_internal(nth_root_mont, psi_pow);
let mut den: Montgomery<u64> = self.montgomery.mul_internal::<ONCE>(nth_root_mont, psi_pow);
den = self.montgomery.pow(den, self.phi-1);
psi_mont = self.montgomery.add_internal(psi_mont, self.montgomery.mul_internal(num, den));
psi_mont = self.montgomery.add_internal(psi_mont, self.montgomery.mul_internal::<ONCE>(num, den));
}
self.montgomery.unprepare(psi_mont)
self.montgomery.unprepare::<ONCE>(psi_mont)
}
}
@@ -201,17 +200,17 @@ impl Prime<u64>{
pub fn Pow(x:u64, exponent:u64, q:u64) -> u64{
let montgomery: MontgomeryPrecomp<u64> = MontgomeryPrecomp::<u64>::new(q);
let mut y_mont: Montgomery<u64> = montgomery.one();
let mut x_mont: Montgomery<u64> = montgomery.prepare(x);
let mut x_mont: Montgomery<u64> = montgomery.prepare::<ONCE>(x);
let mut i: u64 = exponent;
while i > 0{
if i & 1 == 1{
montgomery.mul_internal_assign(x_mont, &mut y_mont);
montgomery.mul_internal_assign::<ONCE>(x_mont, &mut y_mont);
}
montgomery.mul_internal_assign(x_mont, &mut x_mont);
montgomery.mul_internal_assign::<ONCE>(x_mont, &mut x_mont);
i >>= 1;
}
montgomery.unprepare(y_mont)
montgomery.unprepare::<ONCE>(y_mont)
}

11
src/modulus/montgomery.rs
View File

@@ -23,11 +23,22 @@ impl<O> Montgomery<O>{
}
}
/// Default instantiation.
impl<O> Default for Montgomery<O> where O:Default {
fn default() -> Self {
Self {
0: O::default(),
}
}
}
/// MontgomeryPrecomp is a generic struct storing
/// precomputations for Montgomery arithmetic.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub struct MontgomeryPrecomp<O>{
pub q: O,
pub two_q: O,
pub four_q: O,
pub barrett: BarrettPrecomp<O>,
pub q_inv: O,
pub one: Montgomery<O>,

12
src/poly.rs
View File

@@ -3,16 +3,22 @@ pub mod poly;
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct Poly<O>(pub Vec<O>);
impl Poly<u64>{
impl<O> Poly<O>where
O: Default + Clone,
{
pub fn new(n: usize) -> Self{
Self(vec![0u64;n])
Self(vec![O::default();n])
}
pub fn new_montgomery(n: usize) -> Self{
Self(vec![O::default();n])
}
pub fn buffer_size(&self) -> usize{
return self.0.len()
}
pub fn from_buffer(&mut self, n: usize, buf: &mut [u64]){
pub fn from_buffer(&mut self, n: usize, buf: &mut [O]){
assert!(buf.len() >= n, "invalid buffer: buf.len()={} < n={}", buf.len(), n);
self.0 = Vec::from(&buf[..n]);
}

11
src/ring/impl_u64/ring.rs
View File

@@ -1,6 +1,7 @@
use crate::ring::Ring;
use crate::dft::ntt::Table;
use crate::modulus::prime::Prime;
use crate::modulus::montgomery::Montgomery;
use crate::poly::Poly;
impl Ring<u64>{
@@ -17,7 +18,15 @@ impl Ring<u64>{
return self.n
}
fn new_poly_core<O>(&self) -> Poly<O> where O: Default + Clone {
Poly::<O>::new(self.n())
}
pub fn new_poly(&self) -> Poly<u64>{
Poly::<u64>::new(self.n())
self.new_poly_core::<u64>()
}
pub fn new_poly_montgomery(&self) -> Poly<Montgomery<u64>>{
self.new_poly_core::<Montgomery<u64>>()
}
}