diff --git a/sm4/gcm_amd64.s b/sm4/gcm_amd64.s
index 9835759..d964838 100644
--- a/sm4/gcm_amd64.s
+++ b/sm4/gcm_amd64.s
@@ -297,21 +297,19 @@ TEXT ·gcmSm4Init(SB),NOSPLIT,$0
 	XORL CX, CX
 
 sm4InitEncLoop:
-	SM4_SINGLE_ROUND(0, RK, CX, T0, T1, T2, B0, B1, B2, B3)
-	SM4_SINGLE_ROUND(1, RK, CX, T0, T1, T2, B1, B2, B3, B0)
-	SM4_SINGLE_ROUND(2, RK, CX, T0, T1, T2, B2, B3, B0, B1)
-	SM4_SINGLE_ROUND(3, RK, CX, T0, T1, T2, B3, B0, B1, B2)
+	SM4_SINGLE_ROUND(0, RK, CX, T0, T1, T2, B3, B2, B1, B0)
+	SM4_SINGLE_ROUND(1, RK, CX, T0, T1, T2, B2, B1, B0, B3)
+	SM4_SINGLE_ROUND(2, RK, CX, T0, T1, T2, B1, B0, B3, B2)
+	SM4_SINGLE_ROUND(3, RK, CX, T0, T1, T2, B0, B3, B2, B1)
 
 	ADDL $16, CX
 	CMPL CX, $4*32
 	JB sm4InitEncLoop
 
-	PEXTRD $0, B1, R8
-	PINSRD $1, R8, B0
-	PEXTRD $0, B2, R8
-	PINSRD $2, R8, B0
-	PEXTRD $0, B3, R8
-	PINSRD $3, R8, B0
+	PALIGNR $4, B3, B3
+	PALIGNR $4, B3, B2
+	PALIGNR $4, B2, B1
+	PALIGNR $4, B1, B0
 
 	// H * 2
 	PSHUFD $0xff, B0, T0