use std::sync::atomic::AtomicU64;
use crate::algorithms;
use crate::computation::*;
use crate::divisibility::*;
use crate::homomorphism::Homomorphism;
use crate::ordered::OrderedRingStore;
use crate::primitive_int::StaticRing;
use crate::ring::*;
use crate::rings::finite::*;
use crate::integer::*;
use crate::rings::zn::*;
use crate::pid::PrincipalIdealRingStore;
use crate::algorithms::eea::signed_gcd;
use crate::algorithms::sqr_mul;
use crate::seq::VectorFn;
use crate::MAX_PROBABILISTIC_REPETITIONS;
use super::int_factor::is_prime_power;
type Point<R> = (El<R>, El<R>, El<R>);
fn square<R>(Zn: &R, x: &El<R>) -> El<R>
where R: RingStore
{
let mut result: <<R as RingStore>::Type as RingBase>::Element = Zn.clone_el(&x);
Zn.square(&mut result);
return result;
}
#[allow(unused)]
fn point_eq<R>(Zn: &R, P: &Point<R>, Q: &Point<R>) -> bool
where R: RingStore,
R::Type: ZnRing
{
let factor_quo = if !Zn.is_zero(&Q.0) {
if Zn.is_zero(&P.0) { return false; }
(&P.0, &Q.0)
} else if !Zn.is_zero(&Q.1) {
if Zn.is_zero(&P.1) { return false; }
(&P.1, &Q.1)
} else {
assert!(!Zn.is_zero(&Q.2));
if Zn.is_zero(&P.2) { return false; }
(&P.2, &Q.2)
};
if !Zn.is_unit(&factor_quo.1) {
let factor_of_n = signed_gcd(Zn.integer_ring().clone_el(Zn.modulus()), Zn.smallest_positive_lift(Zn.clone_el(&factor_quo.1)), Zn.integer_ring());
let Zn_new = zn_big::Zn::new(BigIntRing::RING, int_cast(Zn.integer_ring().checked_div(Zn.modulus(), &factor_of_n).unwrap(), BigIntRing::RING, Zn.integer_ring()));
let red_map = ZnReductionMap::new(Zn, &Zn_new).unwrap();
if (Zn_new.is_zero(&red_map.map_ref(&Q.0)) && Zn_new.is_zero(&red_map.map_ref(&Q.1)) && Zn_new.is_zero(&red_map.map_ref(&Q.2))) || (Zn_new.is_zero(&red_map.map_ref(&P.0)) && Zn_new.is_zero(&red_map.map_ref(&P.1)) && Zn_new.is_zero(&red_map.map_ref(&P.2))) {
if (Zn_new.is_zero(&red_map.map_ref(&P.0)) && Zn_new.is_zero(&red_map.map_ref(&P.1)) && Zn_new.is_zero(&red_map.map_ref(&P.2))) != (Zn_new.is_zero(&red_map.map_ref(&Q.0)) && Zn_new.is_zero(&red_map.map_ref(&Q.1)) && Zn_new.is_zero(&red_map.map_ref(&Q.2))) {
return false;
}
} else if !point_eq(&Zn_new, &(red_map.map_ref(&P.0), red_map.map_ref(&P.1), red_map.map_ref(&P.2)), &(red_map.map_ref(&Q.0), red_map.map_ref(&Q.1), red_map.map_ref(&Q.2))) {
return false;
}
let Zn_new = zn_big::Zn::new(BigIntRing::RING, int_cast(factor_of_n, BigIntRing::RING, Zn.integer_ring()));
let red_map = ZnReductionMap::new(Zn, &Zn_new).unwrap();
if (Zn_new.is_zero(&red_map.map_ref(&Q.0)) && Zn_new.is_zero(&red_map.map_ref(&Q.1)) && Zn_new.is_zero(&red_map.map_ref(&Q.2))) || (Zn_new.is_zero(&red_map.map_ref(&P.0)) && Zn_new.is_zero(&red_map.map_ref(&P.1)) && Zn_new.is_zero(&red_map.map_ref(&P.2))) {
if (Zn_new.is_zero(&red_map.map_ref(&P.0)) && Zn_new.is_zero(&red_map.map_ref(&P.1)) && Zn_new.is_zero(&red_map.map_ref(&P.2))) != (Zn_new.is_zero(&red_map.map_ref(&Q.0)) && Zn_new.is_zero(&red_map.map_ref(&Q.1)) && Zn_new.is_zero(&red_map.map_ref(&Q.2))) {
return false;
}
} else if !point_eq(&Zn_new, &(red_map.map_ref(&P.0), red_map.map_ref(&P.1), red_map.map_ref(&P.2)), &(red_map.map_ref(&Q.0), red_map.map_ref(&Q.1), red_map.map_ref(&Q.2))) {
return false;
}
return true;
}
let factor = Zn.checked_div(&factor_quo.0, &factor_quo.1).unwrap();
if !Zn.is_unit(&factor) {
return false;
}
return Zn.eq_el(&P.0, &Zn.mul_ref(&factor, &Q.0)) && Zn.eq_el(&P.1, &Zn.mul_ref(&factor, &Q.1)) && Zn.eq_el(&P.2, &Zn.mul_ref(&factor, &Q.2));
}
#[inline(never)]
fn edcurve_add<R>(Zn: &R, d: &El<R>, P: Point<R>, Q: &Point<R>) -> Point<R>
where R: RingStore,
R::Type: ZnRing
{
let (Px, Py, Pz) = P;
let (Qx, Qy, Qz) = Q;
let PxQx = Zn.mul_ref(&Px, Qx);
let PyQy = Zn.mul_ref(&Py, Qy);
let PzQz = Zn.mul_ref_snd(Pz, Qz);
let PzQz_sqr = square(Zn, &PzQz);
let dPxPyQxQy = Zn.mul_ref_snd(Zn.mul_ref(&PxQx, &PyQy), d);
let u1 = Zn.add_ref(&PzQz_sqr, &dPxPyQxQy);
let u2 = Zn.sub(PzQz_sqr, dPxPyQxQy);
let result = (
Zn.mul_ref_fst(&PzQz, Zn.mul_ref_snd(Zn.add(Zn.mul_ref_snd(Px, Qy), Zn.mul_ref_snd(Py, Qx)), &u2)),
Zn.mul(PzQz, Zn.mul_ref_snd(Zn.sub(PyQy, PxQx), &u1)),
Zn.mul(u1, u2),
);
debug_assert!(is_on_curve(Zn, d, &result));
return result;
}
#[inline(never)]
fn edcurve_double<R>(Zn: &R, d: &El<R>, P: &Point<R>) -> Point<R>
where R: RingStore,
R::Type: ZnRing
{
let (Px, Py, Pz) = P;
let PxPy = Zn.mul_ref(&Px, &Py);
let Px_sqr = square(Zn, Px);
let Py_sqr = square(Zn, Py);
let Pz_sqr = square(Zn, Pz);
let Pz_pow4 = square(Zn, &Pz_sqr);
let d_PxPy_sqr = Zn.mul_ref_snd(Zn.mul_ref(&Px_sqr, &Py_sqr), d);
let u1 = Zn.add_ref(&Pz_pow4, &d_PxPy_sqr);
let u2 = Zn.sub(Pz_pow4, d_PxPy_sqr);
let result = (
Zn.mul_ref_fst(&Pz_sqr, Zn.mul_ref_snd(Zn.add_ref(&PxPy, &PxPy), &u2)),
Zn.mul_ref_fst(&Pz_sqr, Zn.mul_ref_snd(Zn.sub(Py_sqr, Px_sqr), &u1)),
Zn.mul(u1, u2),
);
debug_assert!(is_on_curve(Zn, d, &result));
return result;
}
fn ec_pow<R>(base: &Point<R>, d: &El<R>, power: &El<BigIntRing>, Zn: &R) -> Point<R>
where R: RingStore,
R::Type: ZnRing
{
let copy_point = |(x, y, z): &Point<R>| (Zn.clone_el(x), Zn.clone_el(y), Zn.clone_el(z));
let ZZ = BigIntRing::RING;
sqr_mul::generic_pow_shortest_chain_table(
copy_point(base),
power,
ZZ,
|P| Ok(edcurve_double(Zn, d, &P)),
|P, Q| Ok(edcurve_add(Zn, d, copy_point(Q), P)),
|P| copy_point(P),
(Zn.zero(), Zn.one(), Zn.one())
).unwrap_or_else(|x| x)
}
fn is_on_curve<R>(Zn: &R, d: &El<R>, P: &Point<R>) -> bool
where R: RingStore,
R::Type: ZnRing
{
let (x, y, z) = &P;
let x_sqr = square(Zn, x);
let y_sqr = square(Zn, y);
let z_sqr = square(Zn, z);
Zn.eq_el(
&Zn.mul_ref_snd(Zn.add_ref(&x_sqr, &y_sqr), &z_sqr),
&Zn.add(
Zn.mul_ref(&z_sqr, &z_sqr),
Zn.mul_ref_fst(d, Zn.mul(x_sqr, y_sqr))
)
)
}
const POW_COST_CONSTANT: f64 = 0.1;
fn optimize_parameters(ln_p: f64, ln_n: f64) -> (f64, f64) {
let pow_cost_constant = POW_COST_CONSTANT;
let ln_cost_per_attempt = |ln_B: f64| ln_B + ln_B.ln() + pow_cost_constant * ln_n.ln();
let ln_cost_per_attempt_diff = |ln_B: f64| 1. + 1./ln_B;
let ln_attempts = |ln_B: f64| {
let u = ln_p / ln_B;
u * (1. + 2f64.ln()) * u.ln() - u
};
let ln_attempts_diff = |ln_B: f64| {
let u = ln_p / ln_B;
let u_diff = -ln_p / (ln_B * ln_B);
u_diff * (1. + 2f64.ln()) * u.ln() + u * (1. + 2f64.ln()) * u_diff/u - u_diff
};
let f = |ln_B: f64| ln_cost_per_attempt(ln_B) - ln_attempts(ln_B);
let f_diff = |ln_B: f64| ln_cost_per_attempt_diff(ln_B) - ln_attempts_diff(ln_B);
let mut ln_B = (ln_p * ln_p.ln()).sqrt();
for _ in 0..10 {
ln_B = ln_B - f(ln_B) / f_diff(ln_B);
}
return (ln_B, ln_attempts(ln_B));
}
fn lenstra_ec_factor_base<R, F, Controller>(Zn: R, log2_p: usize, mut rng: F, controller: Controller) -> Result<Option<El<<R::Type as ZnRing>::IntegerRing>>, Controller::Abort>
where R: RingStore + Copy + Send + Sync,
El<R>: Send,
R::Type: ZnRing + DivisibilityRing,
F: FnMut() -> u64 + Send,
Controller: ComputationController
{
controller.run_computation(format_args!("ec_factor(log2(n)={}, log2(p)={})", Zn.integer_ring().abs_log2_ceil(Zn.modulus()).unwrap(), log2_p), |controller| {
let ZZ = BigIntRing::RING;
assert!(ZZ.is_leq(&ZZ.power_of_two(log2_p * 2), &Zn.size(&ZZ).unwrap()));
let log2_n = ZZ.abs_log2_ceil(&Zn.size(&ZZ).unwrap()).unwrap();
let ln_p = log2_p as f64 * 2f64.ln();
let (ln_B, ln_attempts) = optimize_parameters(ln_p, log2_n as f64 * 2f64.ln());
let attempts = ln_attempts.exp() as usize;
log_progress!(controller, "(attempts={})", attempts);
let log2_B = ln_B / 2f64.ln();
assert!(log2_B <= i128::MAX as f64);
let primes = algorithms::erathostenes::enumerate_primes(&StaticRing::<i128>::RING, &(1i128 << (log2_B as u64)));
let power_factorization = primes.iter()
.map(|p| (*p, log2_B.ceil() as usize / StaticRing::<i128>::RING.abs_log2_ceil(&p).unwrap()))
.collect::<Vec<_>>();
let power = ZZ.prod(power_factorization.iter().map(|(p, e)| ZZ.pow(ZZ.coerce(&StaticRing::<i128>::RING, *p), *e)));
let power_ref = &power;
let computation = ShortCircuitingComputation::new();
let base_rng_value = rng();
let rng_seed = AtomicU64::new(1);
let rng_seed_ref = &rng_seed;
computation.handle(controller.clone()).join_many((0..attempts).map_fn(move |_| move |handle: ShortCircuitingComputationHandle<_, _>| {
let mut rng = oorandom::Rand64::new(((rng_seed_ref.fetch_add(1, std::sync::atomic::Ordering::Relaxed) as u128) << 64) | base_rng_value as u128);
let (x, y) = (Zn.random_element(|| rng.rand_u64()), Zn.random_element(|| rng.rand_u64()));
let (x_sqr, y_sqr) = (square(&Zn, &x), square(&Zn, &y));
if let Some(d) = Zn.checked_div(&Zn.sub(Zn.add_ref(&x_sqr, &y_sqr), Zn.one()), &Zn.mul(x_sqr, y_sqr)) {
let P = (x, y, Zn.one());
debug_assert!(is_on_curve(&Zn, &d, &P));
let result = ec_pow(&P, &d, power_ref, &Zn).0;
if !Zn.is_unit(&result) && !Zn.is_zero(&result) {
return Ok(Some(result));
}
}
log_progress!(handle, ".");
checkpoint!(handle);
return Ok(None);
}));
if let Some(result) = computation.finish()? {
return Ok(Some(Zn.integer_ring().ideal_gen(&Zn.smallest_positive_lift(result), Zn.modulus())));
} else {
log_progress!(controller, "(no_factor)");
return Ok(None);
}
})
}
#[stability::unstable(feature = "enable")]
pub fn lenstra_ec_factor_small<R, Controller>(Zn: R, min_factor_bound_log2: usize, repetitions: usize, controller: Controller) -> Result<Option<El<<R::Type as ZnRing>::IntegerRing>>, Controller::Abort>
where R: ZnRingStore + DivisibilityRingStore + Copy + Send + Sync,
El<R>: Send,
R::Type: ZnRing + DivisibilityRing,
Controller: ComputationController
{
assert!(algorithms::miller_rabin::is_prime_base(&Zn, 10) == false);
assert!(is_prime_power(Zn.integer_ring(), Zn.modulus()).is_none());
let mut rng = oorandom::Rand64::new(Zn.integer_ring().default_hash(Zn.modulus()) as u128);
for log2_size in (16..min_factor_bound_log2).step_by(8) {
if let Some(factor) = lenstra_ec_factor_base(Zn, log2_size, || rng.rand_u64(), controller.clone())? {
return Ok(Some(factor));
}
}
for _ in 0..repetitions {
if let Some(factor) = lenstra_ec_factor_base(Zn, min_factor_bound_log2, || rng.rand_u64(), controller.clone())? {
return Ok(Some(factor));
}
}
return Ok(None);
}
#[stability::unstable(feature = "enable")]
pub fn lenstra_ec_factor<R, Controller>(Zn: R, controller: Controller) -> Result<El<<R::Type as ZnRing>::IntegerRing>, Controller::Abort>
where R: ZnRingStore + DivisibilityRingStore + Copy + Send + Sync,
El<R>: Send,
R::Type: ZnRing + DivisibilityRing,
Controller: ComputationController
{
assert!(algorithms::miller_rabin::is_prime_base(&Zn, 10) == false);
assert!(is_prime_power(Zn.integer_ring(), Zn.modulus()).is_none());
let ZZ = BigIntRing::RING;
let log2_N = ZZ.abs_log2_floor(&Zn.size(&ZZ).unwrap()).unwrap();
let mut rng = oorandom::Rand64::new(Zn.integer_ring().default_hash(Zn.modulus()) as u128);
for log2_size in (16..(log2_N / 2)).step_by(8) {
if let Some(factor) = lenstra_ec_factor_base(Zn, log2_size, || rng.rand_u64(), controller.clone())? {
return Ok(factor);
}
}
for _ in 0..MAX_PROBABILISTIC_REPETITIONS {
if let Some(factor) = lenstra_ec_factor_base(Zn, log2_N / 2, || rng.rand_u64(), controller.clone())? {
return Ok(factor);
}
}
unreachable!()
}
#[cfg(test)]
use crate::rings::zn::zn_64::Zn;
#[cfg(test)]
use std::time::Instant;
#[cfg(test)]
use test::Bencher;
#[cfg(test)]
use crate::rings::rust_bigint::*;
#[test]
fn test_ec_factor() {
let n = 65537 * 65539;
let actual = lenstra_ec_factor(&Zn::new(n as u64), TEST_LOG_PROGRESS).unwrap_or_else(no_error);
assert!(actual != 1 && actual != n && n % actual == 0);
}
#[bench]
fn bench_ec_factor_mersenne_number_58(bencher: &mut Bencher) {
let bits = 58;
let n = ((1i64 << bits) + 1) / 5;
let ring = Zn::new(n as u64);
bencher.iter(|| {
let p = lenstra_ec_factor(&ring, TEST_LOG_PROGRESS).unwrap_or_else(no_error);
assert!(n > 0 && n != 1 && n != p);
assert!(n % p == 0);
});
}
#[test]
#[ignore]
fn test_ec_factor_large() {
let ZZbig = BigIntRing::RING;
#[cfg(not(feature = "parallel"))]
let controller = TEST_LOG_PROGRESS;
#[cfg(feature = "parallel")]
let controller = RunMultithreadedLogProgress;
let n: i128 = 1073741827 * 71316922984999;
let begin = Instant::now();
let p = StaticRing::<i128>::RING.coerce(&ZZbig, lenstra_ec_factor(&zn_big::Zn::new(&ZZbig, ZZbig.coerce(&StaticRing::<i128>::RING, n)), controller.clone()).unwrap_or_else(no_error));
let end = Instant::now();
println!("Done in {} ms", (end - begin).as_millis());
assert!(p == 1073741827 || p == 71316922984999);
let n: i128 = 1152921504606847009 * 2305843009213693967;
let begin = Instant::now();
let p = StaticRing::<i128>::RING.coerce(&ZZbig, lenstra_ec_factor(&zn_big::Zn::new(&ZZbig, ZZbig.coerce(&StaticRing::<i128>::RING, n)), controller).unwrap_or_else(no_error));
let end = Instant::now();
println!("Done in {} ms", (end - begin).as_millis());
assert!(p == 1152921504606847009 || p == 2305843009213693967);
}
#[test]
#[ignore]
fn test_compute_partial_factorization() {
let ZZbig = BigIntRing::RING;
let n = int_cast(
RustBigintRing::RING.get_ring().parse("5164499756173817179311838344006023748659411585658447025661318713081295244033682389259290706560275662871806343945494986751", 10).unwrap(),
ZZbig,
RustBigintRing::RING
);
let Zn = zn_big::Zn::new(ZZbig, ZZbig.clone_el(&n));
let begin = Instant::now();
let factor = lenstra_ec_factor_small(&Zn, 50, 1, TEST_LOG_PROGRESS).unwrap_or_else(no_error).unwrap();
let end = Instant::now();
println!("Done in {} ms", (end - begin).as_millis());
ZZbig.println(&factor);
assert!(!ZZbig.is_one(&factor));
assert!(!ZZbig.eq_el(&factor, &n));
assert!(ZZbig.divides(&n, &factor));
}