diff --git a/README-CN.md b/README-CN.md
index 2939d0e..e8bd6c0 100644
--- a/README-CN.md
+++ b/README-CN.md
@@ -4,6 +4,7 @@
 使用SIMD指令加速的Base64实现。
 
 [![ci](https://github.com/emmansun/base64/actions/workflows/ci.yml/badge.svg)](https://github.com/emmansun/base64/actions/workflows/ci.yml)
+[![arm64-qemu](https://github.com/emmansun/base64/actions/workflows/ci_qemu.yml/badge.svg)](https://github.com/emmansun/base64/actions/workflows/ci_qemu.yml)
 ![GitHub go.mod Go version (branch)](https://img.shields.io/github/go-mod/go-version/emmansun/base64)
 
 ## 致谢
diff --git a/README.md b/README.md
index b4bf1e9..a7a47dc 100644
--- a/README.md
+++ b/README.md
@@ -4,6 +4,7 @@ English | [简体中文](README-CN.md)
 Base64 with SIMD acceleration
 
 [![ci](https://github.com/emmansun/base64/actions/workflows/ci.yml/badge.svg)](https://github.com/emmansun/base64/actions/workflows/ci.yml)
+[![arm64-qemu](https://github.com/emmansun/base64/actions/workflows/ci_qemu.yml/badge.svg)](https://github.com/emmansun/base64/actions/workflows/ci_qemu.yml)
 ![GitHub go.mod Go version (branch)](https://img.shields.io/github/go-mod/go-version/emmansun/base64)
 
 ## Acknowledgements
diff --git a/base64_arm64.go b/base64_arm64.go
new file mode 100644
index 0000000..cd9b8f3
--- /dev/null
+++ b/base64_arm64.go
@@ -0,0 +1,7 @@
+//go:build arm64 && !purego
+// +build arm64,!purego
+
+package base64
+
+//go:noescape
+func encodeAsm(dst, src []byte, lut *[64]byte) int
diff --git a/base64_arm64.s b/base64_arm64.s
new file mode 100644
index 0000000..755c34a
--- /dev/null
+++ b/base64_arm64.s
@@ -0,0 +1,53 @@
+// Reference
+// https://github.com/aklomp/base64/blob/master/lib/arch/neon64/enc_loop.c
+//go:build arm64 && !purego
+// +build arm64,!purego
+
+//func encodeAsm(dst, src []byte, lut *[64]byte) int
+TEXT ·encodeAsm(SB),NOSPLIT,$0
+	MOVD dst_base+0(FP), R0
+	MOVD src_base+24(FP), R1
+	MOVD src_len+32(FP), R2
+	MOVD lut+48(FP), R3
+
+	VLD1 (R3), [V8.B16, V9.B16, V10.B16, V11.B16]
+	MOVB $0x3F, R4
+	VDUP R3, V7.B16
+	EOR R5, R5, R5
+
+loop:
+	CMP $48, R2
+	BLT done
+
+	// Move the input bits to where they need to be in the outputs. Except
+	// for the first output, the high two bits are not cleared.
+	VLD3.P 48(R1), [V0.B16, V1.B16, V2.B16]
+	VUSHR $2, V0.B16, V3.B16
+	VUSHR $4, V1.B16, V4.B16
+	VUSHR $6, V2.B16, V5.B16
+	VSLI $4, V0.B16, V4.B16
+	VSLI $2, V1.B16, V5.B16
+
+	// Clear the high two bits in the second, third and fourth output.
+	VAND V7.B16, V4.B16, V4.B16
+	VAND V7.B16, V5.B16, V5.B16
+	VAND V7.B16, V2.B16, V6.B16
+
+	// The bits have now been shifted to the right locations;
+	// translate their values 0..63 to the Base64 alphabet.
+	// Use a 64-byte table lookup:
+	VTBL V3.B16, [V8.B16, V9.B16, V10.B16, V11.B16], V3.B16
+	VTBL V4.B16, [V8.B16, V9.B16, V10.B16, V11.B16], V4.B16
+	VTBL V5.B16, [V8.B16, V9.B16, V10.B16, V11.B16], V5.B16
+	VTBL V6.B16, [V8.B16, V9.B16, V10.B16, V11.B16], V6.B16
+
+	// Interleave and store output:
+	VST4.P [V3.B16, V4.B16, V5.B16, V6.B16], 64(R0)
+
+	SUB $48, R2
+	ADD $64, R5
+	B loop
+
+done:
+	MOVD R5, ret+56(FP)
+	RET
diff --git a/base64_arm64_test.go b/base64_arm64_test.go
new file mode 100644
index 0000000..795c54a
--- /dev/null
+++ b/base64_arm64_test.go
@@ -0,0 +1,25 @@
+//go:build arm64 && !purego
+// +build arm64,!purego
+
+package base64
+
+func TestStdEncodeSIMD(t *testing.T) {
+	pairs := []testpair{
+		{"abcdefghijklabcdefghijklabcdefghijklabcdefghijkl", "YWJjZGVmZ2hpamtsYWJjZGVmZ2hpamtsYWJjZGVmZ2hpamtsYWJjZGVmZ2hpamts"},
+		{"abcdefghijklabcdefghijklabcdefghijklabcdefghijklabcdefghijklabcdefghijklabcdefghijklabcdefghijkl", "YWJjZGVmZ2hpamtsYWJjZGVmZ2hpamtsYWJjZGVmZ2hpamtsYWJjZGVmZ2hpamtsYWJjZGVmZ2hpamtsYWJjZGVmZ2hpamtsYWJjZGVmZ2hpamtsYWJjZGVmZ2hpamts"},
+	}
+	for _, p := range pairs {
+		src := []byte(p.decoded)
+		expected := []byte(p.encoded)
+		dst := make([]byte, len(expected))
+
+		ret := encodeAsm(dst, src, &encodeStd)
+		if ret != len(expected) {
+			t.Fatalf("should return %v", len(expected))
+		}
+		if !bytes.Equal(dst, expected) {
+			t.Fatalf("got %v", string(dst))
+		}
+
+	}
+}