diff --git a/sm4/asm_arm64.s b/sm4/asm_arm64.s
index 38d31cb..4ef5e1a 100644
--- a/sm4/asm_arm64.s
+++ b/sm4/asm_arm64.s
@@ -121,7 +121,13 @@ GLOBL fk_mask<>(SB), (NOPTR+RODATA), $16
   VEOR y.B16, x.B16, x.B16
 
 #define SM4_TAO_L2(x, y)         \
-  SM4_SBOX(x, y);                              \
+  LDP	nibble_mask<>(SB), (R0, R1);               \
+	VMOV	R0, XTMP6.D[0];                          \
+	VMOV	R1, XTMP6.D[1];                          \
+  VAND x.B16, XTMP6.B16, XTMP7.B16;              \
+  LDP	m1_low<>(SB), (R0, R1);                    \
+	VMOV	R0, y.D[0];                              \
+	VMOV	R1, y.D[1];                              \
   ;                                           \ //####################  4 parallel L2 linear transforms ##################//
   VSHL $13, x.S4, XTMP6.S4;                   \
   VUSHR $19, x.S4, y.S4;                      \