diff --git a/sm4/asm_arm64.s b/sm4/asm_arm64.s
index 33b219a..aca2c26 100644
--- a/sm4/asm_arm64.s
+++ b/sm4/asm_arm64.s
@@ -108,7 +108,9 @@
 	WORD $0xcec084a8            \ //SM4E V8.4S, V5.4S
 	WORD $0xcec084c8            \ //SM4E V8.4S, V6.4S
 	WORD $0xcec084e8            \ //SM4E V8.4S, V7.4S
-	VREV32 V8.B16, V8.B16       \
+	VREV64	V8.S4, V8.S4               \ 
+	VEXT $8, V8.B16, V8.B16, V8.B16    \	
+	VREV32 V8.B16, V8.B16              \
 	VST1.P  [V8.B16], 16(R9)
 
 // func expandKeyAsm(key *byte, ck, enc, dec *uint32, inst int)
@@ -277,6 +279,8 @@ sm4niblock:
 	WORD $0xcec08428          //SM4E V8.4S, V1.4S
 	WORD $0xcec08448          //SM4E V8.4S, V2.4S
 	WORD $0xcec08468          //SM4E V8.4S, V3.4S
+	VREV64	V8.S4, V8.S4
+	VEXT $8, V8.B16, V8.B16, V8.B16
 	VREV32 V8.B16, V8.B16
 	VST1	[V8.B16], (R9)
 	RET  
diff --git a/sm4/cipher_asm_fuzzy_test.go b/sm4/cipher_asm_fuzzy_test.go
index 23e1622..4bf9987 100644
--- a/sm4/cipher_asm_fuzzy_test.go
+++ b/sm4/cipher_asm_fuzzy_test.go
@@ -36,10 +36,10 @@ func TestExpandKey(t *testing.T) {
 		expandKeyGo(key, encRes1, decRes1)
 		expandKey(key, encRes2, decRes2)
 		if !reflect.DeepEqual(encRes1, encRes2) {
-			t.Errorf("expected=%v, result=%v\n", encRes1, encRes2)
+			t.Errorf("expected=%x, result=%x\n", encRes1, encRes2)
 		}
 		if !reflect.DeepEqual(decRes1, decRes2) {
-			t.Errorf("expected=%v, result=%v\n", encRes1, encRes2)
+			t.Errorf("expected=%x, result=%x\n", encRes1, encRes2)
 		}
 	}
 }
diff --git a/sm4/cipher_test.go b/sm4/cipher_test.go
index 2a3f4d0..c693c55 100644
--- a/sm4/cipher_test.go
+++ b/sm4/cipher_test.go
@@ -30,11 +30,11 @@ func Test_sample1(t *testing.T) {
 	dst := make([]byte, 16)
 	c.Encrypt(dst, src)
 	if !reflect.DeepEqual(dst, expected) {
-		t.Errorf("expected=%v, result=%v\n", expected, dst)
+		t.Errorf("expected=%x, result=%x\n", expected, dst)
 	}
 	c.Decrypt(dst, expected)
 	if !reflect.DeepEqual(dst, src) {
-		t.Errorf("expected=%v, result=%v\n", src, dst)
+		t.Errorf("expected=%x, result=%x\n", src, dst)
 	}
 }
 
@@ -56,7 +56,7 @@ func Test_sample2(t *testing.T) {
 		c.Encrypt(dst, dst)
 	}
 	if !reflect.DeepEqual(dst, expected) {
-		t.Errorf("expected=%v, result=%v\n", expected, dst)
+		t.Errorf("expected=%x, result=%x\n", expected, dst)
 	}
 }
 
diff --git a/sm4/gen_arm64_ni.go b/sm4/gen_arm64_ni.go
index 0db7724..efe2cd6 100644
--- a/sm4/gen_arm64_ni.go
+++ b/sm4/gen_arm64_ni.go
@@ -10,20 +10,21 @@ import (
 	"bytes"
 	"fmt"
 	"log"
-	"math/bits"
 	"os"
 )
 
 //SM4E <Vd>.4S, <Vn>.4S
 func sm4e(Vd, Vn byte) uint32 {
 	inst := uint32(0xcec08400) | uint32(Vd&0x1f) | uint32(Vn&0x1f)<<5
-	return bits.ReverseBytes32(inst)
+	// return bits.ReverseBytes32(inst)
+	return inst
 }
 
 //SM4EKEY <Vd>.4S, <Vn>.4S, <Vm>.4S
 func sm4ekey(Vd, Vn, Vm byte) uint32 {
 	inst := uint32(0xce60c800) | uint32(Vd&0x1f) | uint32(Vn&0x1f)<<5 | (uint32(Vm&0x1f) << 16)
-	return bits.ReverseBytes32(inst)
+	// return bits.ReverseBytes32(inst)
+	return inst
 }
 
 func sm4ekeyRound(buf *bytes.Buffer, d, n, m byte) {
@@ -93,6 +94,8 @@ TEXT ·encryptBlockSM4E(SB),NOSPLIT,$0
 	sm4eRound(buf, 8, 2)
 	sm4eRound(buf, 8, 3)
 	fmt.Fprintf(buf, `
+	VREV64	V8.S4, V8.S4
+	VEXT $8, V8.B16, V8.B16, V8.B16
 	VREV32 V8.B16, V8.B16
 	VST1	[V8.B16], (R9)
 	RET
@@ -121,6 +124,8 @@ TEXT ·encryptBlocksSM4E(SB),NOSPLIT,$0
 		sm4eRound(buf, 8, 5)
 		sm4eRound(buf, 8, 6)
 		sm4eRound(buf, 8, 7)
+		fmt.Fprintf(buf, "\tVREV64	V8.S4, V8.S4\n")
+		fmt.Fprintf(buf, "\tVEXT $8, V8.B16, V8.B16, V8.B16\n")
 		fmt.Fprintf(buf, "\tVREV32 V8.B16, V8.B16\n")
 		fmt.Fprintf(buf, "\tVST1.P	[V8.B16], 16(R9)\n\n")
 	}