diff --git a/src/math/big/arith.go b/src/math/big/arith.go
index b0885f261fe9ba..e621e928f10b6d 100644
--- a/src/math/big/arith.go
+++ b/src/math/big/arith.go
@@ -214,3 +214,20 @@ func divWVW_g(z []Word, xn Word, x []Word, y Word) (r Word) {
 	}
 	return
 }
+func divWWByInv_g(x1, x0, y Word, inv uint, shift uint) (q, r Word) {
+	if shift != 0 {
+		x1 = (x1<<shift | x0>>(bits.UintSize-shift))
+		x0 <<= shift
+		y <<= shift
+	}
+	qq, rr := bits.DivByInv(uint(x1), uint(x0), uint(y), uint(inv))
+	rr >>= shift
+	return Word(qq), Word(rr)
+}
+func divWVWByInv_g(z []Word, xn Word, x []Word, y Word) (r Word) {
+	inv, shift := bits.GetInvert(uint(y))
+	for i := len(z) - 1; i >= 0; i-- {
+		z[i], r = divWWByInv_g(r, x[i], y, inv, shift)
+	}
+	return r
+}
diff --git a/src/math/big/arith_decl.go b/src/math/big/arith_decl.go
index 41e592334c376e..54a2d739e9407b 100644
--- a/src/math/big/arith_decl.go
+++ b/src/math/big/arith_decl.go
@@ -18,3 +18,11 @@ func shrVU(z, x []Word, s uint) (c Word)
 func mulAddVWW(z, x []Word, y, r Word) (c Word)
 func addMulVVW(z, x []Word, y Word) (c Word)
 func divWVW(z []Word, xn Word, x []Word, y Word) (r Word)
+
+//TODO:implemented divWWByInv and divWVWByInv in arith_$GOARCH.s
+func divWWByInv(x1, x0, y Word, inv uint, shift uint) (q, r Word) {
+	return divWWByInv_g(x1, x0, y, inv, shift)
+}
+func divWVWByInv(z []Word, xn Word, x []Word, y Word) (r Word) {
+	return divWVW_g(z, xn, x, y)
+}
diff --git a/src/math/big/nat.go b/src/math/big/nat.go
index 6a3989bf9d82bf..8b2cce3ac04395 100644
--- a/src/math/big/nat.go
+++ b/src/math/big/nat.go
@@ -646,7 +646,7 @@ func (z nat) divW(x nat, y Word) (q nat, r Word) {
 	}
 	// m > 0
 	z = z.make(m)
-	r = divWVW(z, 0, x, y)
+	r = divWVWByInv(z, 0, x, y)
 	q = z.norm()
 	return
 }
@@ -751,6 +751,7 @@ func (q nat) divBasic(u, v nat) {
 
 	// D2.
 	vn1 := v[n-1]
+	inv,shift:=bits.GetInvert(uint(vn1))
 	for j := m; j >= 0; j-- {
 		// D3.
 		qhat := Word(_M)
@@ -760,7 +761,7 @@ func (q nat) divBasic(u, v nat) {
 		}
 		if ujn != vn1 {
 			var rhat Word
-			qhat, rhat = divWW(ujn, u[j+n-1], vn1)
+			qhat, rhat = divWWByInv(ujn, u[j+n-1], vn1,inv,shift)
 
 			// x1 | x2 = q̂v_{n-2}
 			vn2 := v[n-2]
@@ -1179,7 +1180,7 @@ func (x nat) modW(d Word) (r Word) {
 	// TODO(agl): we don't actually need to store the q value.
 	var q nat
 	q = q.make(len(x))
-	return divWVW(q, 0, x, d)
+	return divWVWByInv(q, 0, x, d)
 }
 
 // random creates a random integer in [0..limit), using the space in z if
diff --git a/src/math/bits/bits.go b/src/math/bits/bits.go
index 879ef2da5414f5..c8c6dde04574d8 100644
--- a/src/math/bits/bits.go
+++ b/src/math/bits/bits.go
@@ -586,3 +586,74 @@ func Rem64(hi, lo, y uint64) uint64 {
 	_, rem := Div64(hi%y, lo, y)
 	return rem
 }
+func GetInvert(d1 uint) (inv uint, shift uint) {
+	nlzx, nlzc := d1, uint(0)
+	const SHIFT_BITS int = 8
+	if uintSize > SHIFT_BITS {
+		for ; (nlzx & (uint(0xff) << (uintSize - 8))) == 0; nlzc += 8 {
+			nlzx <<= SHIFT_BITS
+		}
+	}
+	for ; (nlzx & (uint(1) << (uintSize - 1))) == 0; nlzc++ {
+		nlzx <<= 1
+	}
+	shift = nlzc
+	if shift > 0 {
+		d1 = (d1 << shift)
+	}
+	mask := uint((1 << (UintSize / 2)) - 1)
+	max := ^uint(0)
+	u1 := d1
+	ul := u1 & mask
+	uh := u1 >> (uintSize / 2)
+
+	qh := (u1 ^ max) / uh
+
+	r := ((^u1 - qh*uh) << (uintSize / 2)) | mask
+	p := qh * ul
+
+	if r < p {
+		qh--
+		r += u1
+		if r >= u1 {
+			if r < p {
+				qh--
+				r += u1
+			}
+		}
+	}
+	r -= p
+	p = (r>>(uintSize/2))*qh + r
+	ql := (p >> (uintSize / 2)) + 1
+	r = (r << (uintSize / 2)) + mask - ql*u1
+	if r >= (max & (p << (uintSize / 2))) {
+		ql--
+		r += u1
+	}
+	inv = (qh << (uintSize / 2)) + ql
+	if r >= u1 {
+		inv++
+		r -= u1
+	}
+	return
+}
+func DivByInv(n2, n1, d, inv uint) (q uint, r uint) {
+	q, q0 := Mul(n2, inv)
+	q0, cc := Add(q0, n1, 0)
+	q, cc = Add(q, n2, cc)
+
+	r = n1 - d*q
+	r -= d
+	q++
+
+	if r >= q0 {
+		q--
+		r += d
+	}
+
+	if r >= d {
+		q++
+		r -= d
+	}
+	return
+}
\ No newline at end of file