package bn128
|
|
|
|
import (
|
|
"bytes"
|
|
"math/big"
|
|
)
|
|
|
|
// Fq12 uses the same algorithms than Fq2, but with [2][3][2]*big.Int data structure
|
|
|
|
// Fq12 is Field 12
|
|
type Fq12 struct {
|
|
F Fq6
|
|
Fq2 Fq2
|
|
NonResidue [2]*big.Int
|
|
}
|
|
|
|
// NewFq12 generates a new Fq12
|
|
func NewFq12(f Fq6, fq2 Fq2, nonResidue [2]*big.Int) Fq12 {
|
|
fq12 := Fq12{
|
|
f,
|
|
fq2,
|
|
nonResidue,
|
|
}
|
|
return fq12
|
|
}
|
|
|
|
// Zero returns a Zero value on the Fq12
|
|
func (fq12 Fq12) Zero() [2][3][2]*big.Int {
|
|
return [2][3][2]*big.Int{fq12.F.Zero(), fq12.F.Zero()}
|
|
}
|
|
|
|
// One returns a One value on the Fq12
|
|
func (fq12 Fq12) One() [2][3][2]*big.Int {
|
|
return [2][3][2]*big.Int{fq12.F.One(), fq12.F.One()}
|
|
}
|
|
|
|
func (fq12 Fq12) mulByNonResidue(a [3][2]*big.Int) [3][2]*big.Int {
|
|
return [3][2]*big.Int{
|
|
fq12.Fq2.Mul(fq12.NonResidue, a[2]),
|
|
a[0],
|
|
a[1],
|
|
}
|
|
}
|
|
|
|
// Add performs an addition on the Fq12
|
|
func (fq12 Fq12) Add(a, b [2][3][2]*big.Int) [2][3][2]*big.Int {
|
|
return [2][3][2]*big.Int{
|
|
fq12.F.Add(a[0], b[0]),
|
|
fq12.F.Add(a[1], b[1]),
|
|
}
|
|
}
|
|
|
|
// Double performs a doubling on the Fq12
|
|
func (fq12 Fq12) Double(a [2][3][2]*big.Int) [2][3][2]*big.Int {
|
|
return fq12.Add(a, a)
|
|
}
|
|
|
|
// Sub performs a substraction on the Fq12
|
|
func (fq12 Fq12) Sub(a, b [2][3][2]*big.Int) [2][3][2]*big.Int {
|
|
return [2][3][2]*big.Int{
|
|
fq12.F.Sub(a[0], b[0]),
|
|
fq12.F.Sub(a[1], b[1]),
|
|
}
|
|
}
|
|
|
|
// Neg performs a negation on the Fq12
|
|
func (fq12 Fq12) Neg(a [2][3][2]*big.Int) [2][3][2]*big.Int {
|
|
return fq12.Sub(fq12.Zero(), a)
|
|
}
|
|
|
|
// Mul performs a multiplication on the Fq12
|
|
func (fq12 Fq12) Mul(a, b [2][3][2]*big.Int) [2][3][2]*big.Int {
|
|
// Multiplication and Squaring on Pairing-Friendly [2]*big.Ints.pdf; Section 3 (Karatsuba)
|
|
v0 := fq12.F.Mul(a[0], b[0])
|
|
v1 := fq12.F.Mul(a[1], b[1])
|
|
return [2][3][2]*big.Int{
|
|
fq12.F.Add(v0, fq12.mulByNonResidue(v1)),
|
|
fq12.F.Sub(
|
|
fq12.F.Mul(
|
|
fq12.F.Add(a[0], a[1]),
|
|
fq12.F.Add(b[0], b[1])),
|
|
fq12.F.Add(v0, v1)),
|
|
}
|
|
}
|
|
|
|
func (fq12 Fq12) MulScalar(base [2][3][2]*big.Int, e *big.Int) [2][3][2]*big.Int {
|
|
res := fq12.Zero()
|
|
rem := e
|
|
exp := base
|
|
|
|
for !bytes.Equal(rem.Bytes(), big.NewInt(int64(0)).Bytes()) {
|
|
// if rem % 2 == 1
|
|
if bytes.Equal(new(big.Int).Rem(rem, big.NewInt(int64(2))).Bytes(), big.NewInt(int64(1)).Bytes()) {
|
|
res = fq12.Add(res, exp)
|
|
}
|
|
exp = fq12.Double(exp)
|
|
rem = rem.Rsh(rem, 1) // rem = rem >> 1
|
|
}
|
|
return res
|
|
}
|
|
|
|
// Inverse returns the inverse on the Fq12
|
|
func (fq12 Fq12) Inverse(a [2][3][2]*big.Int) [2][3][2]*big.Int {
|
|
t0 := fq12.F.Square(a[0])
|
|
t1 := fq12.F.Square(a[1])
|
|
t2 := fq12.F.Sub(t0, fq12.mulByNonResidue(t1))
|
|
t3 := fq12.F.Inverse(t2)
|
|
return [2][3][2]*big.Int{
|
|
fq12.F.Mul(a[0], t3),
|
|
fq12.F.Neg(fq12.F.Mul(a[1], t3)),
|
|
}
|
|
}
|
|
|
|
// Div performs a division on the Fq12
|
|
func (fq12 Fq12) Div(a, b [2][3][2]*big.Int) [2][3][2]*big.Int {
|
|
return fq12.Mul(a, fq12.Inverse(b))
|
|
}
|
|
|
|
// Square performs a square operation on the Fq12
|
|
func (fq12 Fq12) Square(a [2][3][2]*big.Int) [2][3][2]*big.Int {
|
|
ab := fq12.F.Mul(a[0], a[1])
|
|
|
|
return [2][3][2]*big.Int{
|
|
fq12.F.Sub(
|
|
fq12.F.Mul(
|
|
fq12.F.Add(a[0], a[1]),
|
|
fq12.F.Add(
|
|
a[0],
|
|
fq12.mulByNonResidue(a[1]))),
|
|
fq12.F.Add(
|
|
ab,
|
|
fq12.mulByNonResidue(ab))),
|
|
fq12.F.Add(ab, ab),
|
|
}
|
|
}
|