use r1cs_core::{ConstraintSystem, SynthesisError};
|
|
|
|
use super::PairingGadget as PG;
|
|
|
|
use crate::{
|
|
fields::{fp::FpGadget, fp12::Fp12Gadget, fp2::Fp2Gadget, FieldGadget},
|
|
groups::bls12::{G1Gadget, G1PreparedGadget, G2Gadget, G2PreparedGadget},
|
|
};
|
|
use algebra::{
|
|
curves::bls12::{Bls12, Bls12Parameters, TwistType},
|
|
fields::BitIterator,
|
|
};
|
|
use core::marker::PhantomData;
|
|
|
|
pub struct PairingGadget<P: Bls12Parameters>(PhantomData<P>);
|
|
|
|
type Fp2G<P> = Fp2Gadget<<P as Bls12Parameters>::Fp2Params, <P as Bls12Parameters>::Fp>;
|
|
|
|
impl<P: Bls12Parameters> PairingGadget<P> {
|
|
// Evaluate the line function at point p.
|
|
fn ell<CS: ConstraintSystem<P::Fp>>(
|
|
mut cs: CS,
|
|
f: &mut Fp12Gadget<P::Fp12Params, P::Fp>,
|
|
coeffs: &(Fp2G<P>, Fp2G<P>),
|
|
p: &G1Gadget<P>,
|
|
) -> Result<(), SynthesisError> {
|
|
let zero = FpGadget::<P::Fp>::zero(cs.ns(|| "fpg zero"))?;
|
|
|
|
match P::TWIST_TYPE {
|
|
TwistType::M => {
|
|
let c0 = coeffs.0.clone();
|
|
let mut c1 = coeffs.1.clone();
|
|
let c2 = Fp2G::<P>::new(p.y.clone(), zero);
|
|
|
|
c1.c0 = c1.c0.mul(cs.ns(|| "mul c1.c0"), &p.x)?;
|
|
c1.c1 = c1.c1.mul(cs.ns(|| "mul c1.c1"), &p.x)?;
|
|
*f = f.mul_by_014(cs.ns(|| "sparse mul f"), &c0, &c1, &c2)?;
|
|
Ok(())
|
|
},
|
|
TwistType::D => {
|
|
let c0 = Fp2G::<P>::new(p.y.clone(), zero);
|
|
let mut c1 = coeffs.0.clone();
|
|
let c2 = coeffs.1.clone();
|
|
|
|
c1.c0 = c1.c0.mul(cs.ns(|| "mul c1.c0"), &p.x)?;
|
|
c1.c1 = c1.c1.mul(cs.ns(|| "mul c1.c1"), &p.x)?;
|
|
*f = f.mul_by_034(cs.ns(|| "sparse mul f"), &c0, &c1, &c2)?;
|
|
Ok(())
|
|
},
|
|
}
|
|
}
|
|
|
|
fn exp_by_x<CS: ConstraintSystem<P::Fp>>(
|
|
mut cs: CS,
|
|
f: &Fp12Gadget<P::Fp12Params, P::Fp>,
|
|
) -> Result<Fp12Gadget<P::Fp12Params, P::Fp>, SynthesisError> {
|
|
let mut result = f.cyclotomic_exp(cs.ns(|| "exp_by_x"), P::X)?;
|
|
if P::X_IS_NEGATIVE {
|
|
result.conjugate_in_place(cs.ns(|| "conjugate"))?;
|
|
}
|
|
Ok(result)
|
|
}
|
|
}
|
|
|
|
impl<P: Bls12Parameters> PG<Bls12<P>, P::Fp> for PairingGadget<P> {
|
|
type G1Gadget = G1Gadget<P>;
|
|
type G2Gadget = G2Gadget<P>;
|
|
type G1PreparedGadget = G1PreparedGadget<P>;
|
|
type G2PreparedGadget = G2PreparedGadget<P>;
|
|
type GTGadget = Fp12Gadget<P::Fp12Params, P::Fp>;
|
|
|
|
fn miller_loop<CS: ConstraintSystem<P::Fp>>(
|
|
mut cs: CS,
|
|
ps: &[Self::G1PreparedGadget],
|
|
qs: &[Self::G2PreparedGadget],
|
|
) -> Result<Self::GTGadget, SynthesisError> {
|
|
let mut pairs = vec![];
|
|
for (p, q) in ps.iter().zip(qs.iter()) {
|
|
pairs.push((p, q.ell_coeffs.iter()));
|
|
}
|
|
let mut f = Self::GTGadget::one(cs.ns(|| "one"))?;
|
|
|
|
for (j, i) in BitIterator::new(P::X).skip(1).enumerate() {
|
|
let mut cs = cs.ns(|| format!("Iteration {}", j));
|
|
f.square_in_place(cs.ns(|| "square"))?;
|
|
|
|
for (k, &mut (p, ref mut coeffs)) in pairs.iter_mut().enumerate() {
|
|
let cs = cs.ns(|| format!("Double input {}", k));
|
|
Self::ell(cs, &mut f, coeffs.next().unwrap(), &p.0)?;
|
|
}
|
|
|
|
if i {
|
|
for (k, &mut (p, ref mut coeffs)) in pairs.iter_mut().enumerate() {
|
|
let cs = cs.ns(|| format!("Addition input {}", k));
|
|
Self::ell(cs, &mut f, &coeffs.next().unwrap(), &p.0)?;
|
|
}
|
|
}
|
|
}
|
|
|
|
if P::X_IS_NEGATIVE {
|
|
f.conjugate_in_place(cs.ns(|| "f conjugate"))?;
|
|
}
|
|
|
|
Ok(f)
|
|
}
|
|
|
|
fn final_exponentiation<CS: ConstraintSystem<P::Fp>>(
|
|
mut cs: CS,
|
|
f: &Self::GTGadget,
|
|
) -> Result<Self::GTGadget, SynthesisError> {
|
|
// Computing the final exponentation following
|
|
// https://eprint.iacr.org/2016/130.pdf.
|
|
// We don't use their "faster" formula because it is difficult to make
|
|
// it work for curves with odd `P::X`.
|
|
// Hence we implement the slower algorithm from Table 1 below.
|
|
|
|
let f1 = f.frobenius_map(cs.ns(|| "frobmap 1"), 6)?;
|
|
|
|
f.inverse(cs.ns(|| "inverse 1")).and_then(|mut f2| {
|
|
// f2 = f^(-1);
|
|
// r = f^(p^6 - 1)
|
|
let mut r = f1;
|
|
r.mul_in_place(cs.ns(|| "r = f1 * f2"), &f2)?;
|
|
|
|
// f2 = f^(p^6 - 1)
|
|
f2 = r.clone();
|
|
// r = f^((p^6 - 1)(p^2))
|
|
r.frobenius_map_in_place(cs.ns(|| "frobenius map 2"), 2)?;
|
|
|
|
// r = f^((p^6 - 1)(p^2) + (p^6 - 1))
|
|
// r = f^((p^6 - 1)(p^2 + 1))
|
|
r.mul_in_place(cs.ns(|| "mul 0"), &f2)?;
|
|
|
|
// Hard part of the final exponentation is below:
|
|
// From https://eprint.iacr.org/2016/130.pdf, Table 1
|
|
let mut y0 = r.cyclotomic_square(cs.ns(|| "cyclotomic_sq 1"))?;
|
|
y0.conjugate_in_place(&mut cs.ns(|| "conjugate 2"))?;
|
|
|
|
let mut y5 = Self::exp_by_x(&mut cs.ns(|| "exp_by_x 1"), &r)?;
|
|
|
|
let mut y1 = y5.cyclotomic_square(&mut cs.ns(|| "square 1"))?;
|
|
let mut y3 = y0.mul(&mut cs.ns(|| "mul 1"), &y5)?;
|
|
y0 = Self::exp_by_x(cs.ns(|| "exp_by_x 2"), &y3)?;
|
|
let y2 = Self::exp_by_x(cs.ns(|| "exp_by_x 3"), &y0)?;
|
|
let mut y4 = Self::exp_by_x(cs.ns(|| "exp_by_x 4"), &y2)?;
|
|
y4.mul_in_place(cs.ns(|| "mul 2"), &y1)?;
|
|
y1 = Self::exp_by_x(cs.ns(|| "exp_by_x 5"), &y4)?;
|
|
y3.conjugate_in_place(cs.ns(|| "conjugate 3"))?;
|
|
y1.mul_in_place(cs.ns(|| "mul 3"), &y3)?;
|
|
y1.mul_in_place(cs.ns(|| "mul 4"), &r)?;
|
|
y3 = r.clone();
|
|
y3.conjugate_in_place(cs.ns(|| "conjugate 4"))?;
|
|
y0.mul_in_place(cs.ns(|| "mul 5"), &r)?;
|
|
y0.frobenius_map_in_place(cs.ns(|| "frobmap 3"), 3)?;
|
|
y4.mul_in_place(cs.ns(|| "mul 6"), &y3)?;
|
|
y4.frobenius_map_in_place(cs.ns(|| "frobmap 4"), 1)?;
|
|
y5.mul_in_place(cs.ns(|| "mul 7"), &y2)?;
|
|
y5.frobenius_map_in_place(cs.ns(|| "frobmap 5"), 2)?;
|
|
y5.mul_in_place(cs.ns(|| "mul 8"), &y0)?;
|
|
y5.mul_in_place(cs.ns(|| "mul 9"), &y4)?;
|
|
y5.mul_in_place(cs.ns(|| "mul 10"), &y1)?;
|
|
Ok(y5)
|
|
})
|
|
}
|
|
|
|
fn prepare_g1<CS: ConstraintSystem<P::Fp>>(
|
|
cs: CS,
|
|
p: &Self::G1Gadget,
|
|
) -> Result<Self::G1PreparedGadget, SynthesisError> {
|
|
Self::G1PreparedGadget::from_affine(cs, p)
|
|
}
|
|
|
|
fn prepare_g2<CS: ConstraintSystem<P::Fp>>(
|
|
cs: CS,
|
|
q: &Self::G2Gadget,
|
|
) -> Result<Self::G2PreparedGadget, SynthesisError> {
|
|
Self::G2PreparedGadget::from_affine(cs, q)
|
|
}
|
|
}
|