From: David Leon Gil <coruus@gmail.com>
Date: Fri, 26 Jun 2015 17:29:45 +0000 (-0700)
Subject: math/big: use optimized formula in ModSqrt for 3 mod 4 primes
X-Git-Tag: go1.6beta1~1234
X-Git-Url: http://www.git.cypherpunks.su/?a=commitdiff_plain;h=ea0491b70a82a207a951f869c7c1e5a52dbf410f;p=gostls13.git

math/big: use optimized formula in ModSqrt for 3 mod 4 primes

For primes which are 3 mod 4, using Tonelli-Shanks is slower
and more complicated than using the identity

     a**((p+1)/4) mod p == sqrt(a)

For 2^450-2^225-1 and 2^10860-2^5430-1, which are 3 mod 4:

BenchmarkModSqrt225_TonelliTri      1000     1135375 ns/op
BenchmarkModSqrt225_3Mod4          10000      156009 ns/op
BenchmarkModSqrt5430_Tonelli           1  3448851386 ns/op
BenchmarkModSqrt5430_3Mod4             2   914616710 ns/op

~2.6x to 7x faster.

Fixes #11437 (which is a prime choice of issues to fix)

Change-Id: I813fb29454160483ec29825469e0370d517850c2
Reviewed-on: https://go-review.googlesource.com/11522
Reviewed-by: Adam Langley <agl@golang.org>
---

diff --git a/src/math/big/int.go b/src/math/big/int.go
index 65334e0ef5..84485a2750 100644
--- a/src/math/big/int.go
+++ b/src/math/big/int.go
@@ -640,23 +640,23 @@ func Jacobi(x, y *Int) int {
 	}
 }
 
-// ModSqrt sets z to a square root of x mod p if such a square root exists, and
-// returns z. The modulus p must be an odd prime. If x is not a square mod p,
-// ModSqrt leaves z unchanged and returns nil. This function panics if p is
-// not an odd integer.
-func (z *Int) ModSqrt(x, p *Int) *Int {
-	switch Jacobi(x, p) {
-	case -1:
-		return nil // x is not a square mod p
-	case 0:
-		return z.SetInt64(0) // sqrt(0) mod p = 0
-	case 1:
-		break
-	}
-	if x.neg || x.Cmp(p) >= 0 { // ensure 0 <= x < p
-		x = new(Int).Mod(x, p)
-	}
+// modSqrt3Mod4 uses the identity
+//      (a^((p+1)/4))^2  mod p
+//   == u^(p+1)          mod p
+//   == u^2              mod p
+// to calculate the square root of any quadratic residue mod p quickly for 3
+// mod 4 primes.
+func (z *Int) modSqrt3Mod4Prime(x, p *Int) *Int {
+	z.Set(p)         // z = p
+	z.Add(z, intOne) // z = p + 1
+	z.Rsh(z, 2)      // z = (p + 1) / 4
+	z.Exp(x, z, p)   // z = x^z mod p
+	return z
+}
 
+// modSqrtTonelliShanks uses the Tonelli-Shanks algorithm to find the square
+// root of a quadratic residue modulo any prime.
+func (z *Int) modSqrtTonelliShanks(x, p *Int) *Int {
 	// Break p-1 into s*2^e such that s is odd.
 	var s Int
 	s.Sub(p, intOne)
@@ -703,6 +703,31 @@ func (z *Int) ModSqrt(x, p *Int) *Int {
 	}
 }
 
+// ModSqrt sets z to a square root of x mod p if such a square root exists, and
+// returns z. The modulus p must be an odd prime. If x is not a square mod p,
+// ModSqrt leaves z unchanged and returns nil. This function panics if p is
+// not an odd integer.
+func (z *Int) ModSqrt(x, p *Int) *Int {
+	switch Jacobi(x, p) {
+	case -1:
+		return nil // x is not a square mod p
+	case 0:
+		return z.SetInt64(0) // sqrt(0) mod p = 0
+	case 1:
+		break
+	}
+	if x.neg || x.Cmp(p) >= 0 { // ensure 0 <= x < p
+		x = new(Int).Mod(x, p)
+	}
+
+	// Check whether p is 3 mod 4, and if so, use the faster algorithm.
+	if len(p.abs) > 0 && p.abs[0]%4 == 3 {
+		return z.modSqrt3Mod4Prime(x, p)
+	}
+	// Otherwise, use Tonelli-Shanks.
+	return z.modSqrtTonelliShanks(x, p)
+}
+
 // Lsh sets z = x << n and returns z.
 func (z *Int) Lsh(x *Int, n uint) *Int {
 	z.abs = z.abs.shl(x.abs, n)
diff --git a/src/math/big/int_test.go b/src/math/big/int_test.go
index 97874626f3..5b80509a31 100644
--- a/src/math/big/int_test.go
+++ b/src/math/big/int_test.go
@@ -1185,6 +1185,53 @@ func BenchmarkBitsetNegOrig(b *testing.B) {
 	}
 }
 
+// tri generates the trinomial 2**(n*2) - 2**n - 1, which is always 3 mod 4 and
+// 7 mod 8, so that 2 is always a quadratic residue.
+func tri(n uint) *Int {
+	x := NewInt(1)
+	x.Lsh(x, n)
+	x2 := new(Int).Lsh(x, n)
+	x2.Sub(x2, x)
+	x2.Sub(x2, intOne)
+	return x2
+}
+
+func BenchmarkModSqrt225_Tonelli(b *testing.B) {
+	p := tri(225)
+	x := NewInt(2)
+	for i := 0; i < b.N; i++ {
+		x.SetUint64(2)
+		x.modSqrtTonelliShanks(x, p)
+	}
+}
+
+func BenchmarkModSqrt224_3Mod4(b *testing.B) {
+	p := tri(225)
+	x := new(Int).SetUint64(2)
+	for i := 0; i < b.N; i++ {
+		x.SetUint64(2)
+		x.modSqrt3Mod4Prime(x, p)
+	}
+}
+
+func BenchmarkModSqrt5430_Tonelli(b *testing.B) {
+	p := tri(5430)
+	x := new(Int).SetUint64(2)
+	for i := 0; i < b.N; i++ {
+		x.SetUint64(2)
+		x.modSqrtTonelliShanks(x, p)
+	}
+}
+
+func BenchmarkModSqrt5430_3Mod4(b *testing.B) {
+	p := tri(5430)
+	x := new(Int).SetUint64(2)
+	for i := 0; i < b.N; i++ {
+		x.SetUint64(2)
+		x.modSqrt3Mod4Prime(x, p)
+	}
+}
+
 func TestBitwise(t *testing.T) {
 	x := new(Int)
 	y := new(Int)