precompute part 2

sm9/README.md

@@ -2,3 +2,39 @@ This part codes mainly refer two projects:
 
 1. [bn256](https://github.com/cloudflare/bn256), 主要是基域运算
 2. [gmssl sm9](https://github.com/guanzhi/GmSSL/blob/develop/src/sm9_alg.c)，主要是2-4-12塔式扩域，以及r-ate等
+
+
+**SM9 Sign Benchmark**
+
+    goos: windows
+    goarch: amd64
+    pkg: github.com/emmansun/gmsm/sm9
+    cpu: Intel(R) Core(TM) i5-9500 CPU @ 3.00GHz
+    BenchmarkSign-6   	    1344	    871597 ns/op	   35870 B/op	    1013 allocs/op
+
+
+**SM9 Verify Benchmark**
+
+    goos: windows
+    goarch: amd64
+    pkg: github.com/emmansun/gmsm/sm9
+    cpu: Intel(R) Core(TM) i5-9500 CPU @ 3.00GHz
+    BenchmarkVerify-6   	     352	   3331673 ns/op	  237676 B/op	    6283 allocs/op
+
+**SM9 Encrypt(XOR) Benchmark**
+
+    goos: windows
+    goarch: amd64
+    pkg: github.com/emmansun/gmsm/sm9
+    cpu: Intel(R) Core(TM) i5-9500 CPU @ 3.00GHz
+    BenchmarkEncrypt-6   	    1120	    971188 ns/op	   38125 B/op	    1036 allocs/op
+
+**SM9 Decrypt(XOR) Benchmark**
+
+    goos: windows
+    goarch: amd64
+    pkg: github.com/emmansun/gmsm/sm9
+    cpu: Intel(R) Core(TM) i5-9500 CPU @ 3.00GHz
+    BenchmarkDecrypt-6   	     507	   2345492 ns/op	  202360 B/op	    5228 allocs/op
+
+To further improve `Verify()/Decrypt()` performance, need to improve `Pair()` method performance.

sm9/gfp12.go

@@ -367,3 +367,11 @@ func (e *gfP12) Conjugate(a *gfP12) *gfP12 {
 	e.x.Conjugate(&a.x)
 	return e
 }
+
+// Select sets q to p1 if cond == 1, and to p2 if cond == 0.
+func (q *gfP12) Select(p1, p2 *gfP12, cond int) *gfP12 {
+	q.x.Select(&p1.x, &p2.x, cond)
+	q.y.Select(&p1.y, &p2.y, cond)
+	q.z.Select(&p1.z, &p2.z, cond)
+	return q
+}

sm9/gfp4.go

@@ -241,3 +241,10 @@ func (e *gfP4) FrobeniusP3(a *gfP4) *gfP4 {
 
 	return e
 }
+
+// Select sets q to p1 if cond == 1, and to p2 if cond == 0.
+func (q *gfP4) Select(p1, p2 *gfP4, cond int) *gfP4 {
+	q.x.Select(&p1.x, &p2.x, cond)
+	q.y.Select(&p1.y, &p2.y, cond)
+	return q
+}

sm9/gt.go

@@ -1,6 +1,7 @@
 package sm9
 
 import (
+	"crypto/subtle"
 	"errors"
 	"io"
 	"math/big"
@@ -85,6 +86,15 @@ func (e *GT) Set(a *GT) *GT {
 	return e
 }
 
+// Set sets e to one and then returns e.
+func (e *GT) SetOne() *GT {
+	if e.p == nil {
+		e.p = &gfP12{}
+	}
+	e.p.SetOne()
+	return e	
+}
+
 // Finalize is a linear function from F_p^12 to GT.
 func (e *GT) Finalize() *GT {
 	ret := finalExponentiation(e.p)
@@ -197,3 +207,21 @@ func (e *GT) Unmarshal(m []byte) ([]byte, error) {
 
 	return m[12*numBytes:], nil
 }
+
+// A gtPointTable holds the first 15 Exp of a value at offset -1, so P
+// is at table[0], P^15 is at table[14], and P^0 is implicitly the identity
+// point.
+type gtTable [15]*GT
+
+// Select selects the n-th multiple of the table base point into p. It works in
+// constant time by iterating over every entry of the table. n must be in [0, 15].
+func (table *gtTable) Select(p *GT, n uint8) {
+	if n >= 16 {
+		panic("sm9: internal error: gtTable called with out-of-bounds value")
+	}
+	p.p.SetOne()
+	for i := uint8(1); i < 16; i++ {
+		cond := subtle.ConstantTimeByteEq(i, n)
+		p.p.Select(table[i-1].p, p.p, cond)
+	}
+}

sm9/sm9.go

@@ -91,17 +91,68 @@ func (pub *SignMasterPublicKey) Pair() *GT {
 	return pub.basePoint
 }
 
+func (pub *SignMasterPublicKey) generatorTable() *[32 * 2]gtTable {
+	pub.tableGenOnce.Do(func() {
+		pub.table = new([32 * 2]gtTable)
+		base := &GT{}
+		base.Set(pub.Pair())
+		for i := 0; i < 32*2; i++ {
+			pub.table[i][0] = &GT{}
+			pub.table[i][0].Set(base)
+			for j := 1; j < 15; j += 2 {
+				pub.table[i][j] = &GT{}
+				pub.table[i][j].p = &gfP12{}
+				pub.table[i][j].p.Square(pub.table[i][j/2].p)
+				pub.table[i][j+1] = &GT{}
+				pub.table[i][j+1].p = &gfP12{}
+				pub.table[i][j+1].Add(pub.table[i][j], base)
+			}
+			base.p.Square(base.p)
+			base.p.Square(base.p)
+			base.p.Square(base.p)
+			base.p.Square(base.p)
+		}
+	})
+	return pub.table
+}
+
+func (pub *SignMasterPublicKey) ScalarBaseMult(r *big.Int) *GT {
+	scalar := normalizeScalar(r.Bytes())
+	tables := pub.generatorTable()
+	// This is also a scalar multiplication with a four-bit window like in
+	// ScalarMult, but in this case the doublings are precomputed. The value
+	// [windowValue]G added at iteration k would normally get doubled
+	// (totIterations-k)×4 times, but with a larger precomputation we can
+	// instead add [2^((totIterations-k)×4)][windowValue]G and avoid the
+	// doublings between iterations.
+	e, t := &GT{}, &GT{}
+	tableIndex := len(tables) - 1
+	e.SetOne()
+	t.SetOne()
+	for _, byte := range scalar {
+		windowValue := byte >> 4
+		tables[tableIndex].Select(t, windowValue)
+		e.Add(e, t)
+		tableIndex--
+		windowValue = byte & 0b1111
+		tables[tableIndex].Select(t, windowValue)
+		e.Add(e, t)
+		tableIndex--
+	}
+	return e
+}
+
 // Sign signs a hash (which should be the result of hashing a larger message)
 // using the user dsa key. It returns the signature as a pair of h and s.
 func Sign(rand io.Reader, priv *SignPrivateKey, hash []byte) (h *big.Int, s *G1, err error) {
-	g := priv.Pair()
 	var r *big.Int
 	for {
 		r, err = randFieldElement(rand)
 		if err != nil {
 			return
 		}
-		w := new(GT).ScalarMult(g, r)
+
+		w := priv.SignMasterPublicKey.ScalarBaseMult(r)
 
 		var buffer []byte
 		buffer = append(buffer, hash...)
@@ -157,9 +208,8 @@ func Verify(pub *SignMasterPublicKey, uid []byte, hid byte, hash []byte, h *big.
 	if !s.p.IsOnCurve() {
 		return false
 	}
-	g := pub.Pair()
 
-	t := new(GT).ScalarMult(g, h)
+	t := pub.ScalarBaseMult(h)
 
 	// user sign public key p generation
 	p := pub.GenerateUserPublicKey(uid, hid)
@@ -227,6 +277,57 @@ func (pub *EncryptMasterPublicKey) Pair() *GT {
 	return pub.basePoint
 }
 
+func (pub *EncryptMasterPublicKey) generatorTable() *[32 * 2]gtTable {
+	pub.tableGenOnce.Do(func() {
+		pub.table = new([32 * 2]gtTable)
+		base := &GT{}
+		base.Set(pub.Pair())
+		for i := 0; i < 32*2; i++ {
+			pub.table[i][0] = &GT{}
+			pub.table[i][0].Set(base)
+			for j := 1; j < 15; j += 2 {
+				pub.table[i][j] = &GT{}
+				pub.table[i][j].p = &gfP12{}
+				pub.table[i][j].p.Square(pub.table[i][j/2].p)
+				pub.table[i][j+1] = &GT{}
+				pub.table[i][j+1].p = &gfP12{}
+				pub.table[i][j+1].Add(pub.table[i][j], base)
+			}
+			base.p.Square(base.p)
+			base.p.Square(base.p)
+			base.p.Square(base.p)
+			base.p.Square(base.p)
+		}
+	})
+	return pub.table
+}
+
+func (pub *EncryptMasterPublicKey) ScalarBaseMult(r *big.Int) *GT {
+	scalar := normalizeScalar(r.Bytes())
+	tables := pub.generatorTable()
+	// This is also a scalar multiplication with a four-bit window like in
+	// ScalarMult, but in this case the doublings are precomputed. The value
+	// [windowValue]G added at iteration k would normally get doubled
+	// (totIterations-k)×4 times, but with a larger precomputation we can
+	// instead add [2^((totIterations-k)×4)][windowValue]G and avoid the
+	// doublings between iterations.
+	e, t := &GT{}, &GT{}
+	tableIndex := len(tables) - 1
+	e.SetOne()
+	t.SetOne()
+	for _, byte := range scalar {
+		windowValue := byte >> 4
+		tables[tableIndex].Select(t, windowValue)
+		e.Add(e, t)
+		tableIndex--
+		windowValue = byte & 0b1111
+		tables[tableIndex].Select(t, windowValue)
+		e.Add(e, t)
+		tableIndex--
+	}
+	return e
+}
+
 // WrappKey generate and wrapp key wtih reciever's uid and system hid
 func WrappKey(rand io.Reader, pub *EncryptMasterPublicKey, uid []byte, hid byte, kLen int) (key []byte, cipher *G1, err error) {
 	q := pub.GenerateUserPublicKey(uid, hid)
@@ -240,8 +341,7 @@ func WrappKey(rand io.Reader, pub *EncryptMasterPublicKey, uid []byte, hid byte,
 
 		cipher = new(G1).ScalarMult(q, r)
 
-		g := pub.Pair()
+		w := pub.ScalarBaseMult(r)
-		w := new(GT).ScalarMult(g, r)
 
 		var buffer []byte
 		buffer = append(buffer, cipher.Marshal()...)

sm9/sm9_key.go

@@ -18,6 +18,8 @@ type SignMasterPublicKey struct {
 	MasterPublicKey *G2
 	pairOnce        sync.Once
 	basePoint       *GT
+	tableGenOnce    sync.Once
+	table           *[32 * 2]gtTable
 }
 
 type SignPrivateKey struct {
@@ -34,6 +36,8 @@ type EncryptMasterPublicKey struct {
 	MasterPublicKey *G1
 	pairOnce        sync.Once
 	basePoint       *GT
+	tableGenOnce    sync.Once
+	table           *[32 * 2]gtTable
 }
 
 type EncryptPrivateKey struct {