diff --git a/sm4/cbc_amd64.s b/sm4/cbc_amd64.s
index 5895413..b415f5a 100644
--- a/sm4/cbc_amd64.s
+++ b/sm4/cbc_amd64.s
@@ -132,7 +132,9 @@ TEXT ·decryptBlocksChain(SB),NOSPLIT,$0
 	JE   avx2Start
 
 	CMPB ·useAVX(SB), $1
-	JE   avxCbcSm4Octets
+	JE   avxStart
+
+	MOVOU -16(DX), X15
 
 cbcSm4Octets:
 	CMPQ DI, $128
@@ -155,7 +157,7 @@ cbcSm4Octets:
 	PXOR -16(DX), XWORD0
 	PXOR 0(DX), XWORD1
 	PXOR 16(DX), XWORD2
-	PXOR 32(DX), XWORD3
+	PXOR 32(DX), XWORD3	
 	PXOR 48(DX), XWORD4
 	PXOR 64(DX), XWORD5
 	PXOR 80(DX), XWORD6
@@ -211,8 +213,6 @@ cbCSm4Single:
 	MOVOU -32(DX), XWORD2
 	MOVOU -16(DX), XWORD3
 
-	MOVOU XWORD0, XWORD4
-
 	SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	PXOR 0(SI), XWORD0
@@ -224,28 +224,24 @@ cbCSm4Single:
 	MOVUPS XWORD1, -48(BX)
 	MOVUPS XWORD2, -32(BX)
 	MOVUPS XWORD3, -16(BX)
-	MOVUPS XWORD4, (SI)
+
 	JMP cbcSm4Done
 
 cbcSm4Single16:
 	MOVOU -16(DX), XWORD0
 
-	MOVOU XWORD0, XWORD4
-
 	SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	PXOR 0(SI), XWORD0
 
 	MOVUPS XWORD0, -16(BX)
-	MOVUPS XWORD4, (SI)
+
 	JMP cbcSm4Done
 
 cbcSm4Single32:
 	MOVOU -32(DX), XWORD0
 	MOVOU -16(DX), XWORD1
 
-	MOVOU XWORD0, XWORD4
-
 	SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	PXOR 0(SI), XWORD0
@@ -253,7 +249,7 @@ cbcSm4Single32:
 
 	MOVUPS XWORD0, -32(BX)
 	MOVUPS XWORD1, -16(BX)
-	MOVUPS XWORD4, (SI)
+
 	JMP cbcSm4Done
 
 cbcSm4Single48:
@@ -261,8 +257,6 @@ cbcSm4Single48:
 	MOVOU -32(DX), XWORD1
 	MOVOU -16(DX), XWORD2
 
-	MOVOU XWORD0, XWORD4
-
 	SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	PXOR 0(SI), XWORD0
@@ -272,11 +266,14 @@ cbcSm4Single48:
 	MOVUPS XWORD0, -48(BX)
 	MOVUPS XWORD1, -32(BX)
 	MOVUPS XWORD2, -16(BX)
-	MOVUPS XWORD4, (SI)
 	
 cbcSm4Done:
+	MOVUPS X15, (SI)
 	RET
 
+avxStart:
+	VMOVDQU -16(DX), X15
+
 avxCbcSm4Octets:
 	CMPQ DI, $128
 	JLE avxCbcSm4Nibbles
@@ -354,8 +351,6 @@ avxCbCSm4Single:
 	VMOVDQU -32(DX), XWORD2
 	VMOVDQU -16(DX), XWORD3
 
-	VMOVDQU XWORD0, XWORD4
-
 	AVX_SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	VPXOR 0(SI), XWORD0, XWORD0
@@ -367,28 +362,24 @@ avxCbCSm4Single:
 	VMOVDQU XWORD1, -48(BX)
 	VMOVDQU XWORD2, -32(BX)
 	VMOVDQU XWORD3, -16(BX)
-	VMOVDQU XWORD4, (SI)
+
 	JMP avxCbcSm4Done
 
 avxCbcSm4Single16:
 	VMOVDQU -16(DX), XWORD0
 
-	VMOVDQU XWORD0, XWORD4
-
 	AVX_SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	VPXOR 0(SI), XWORD0, XWORD0
 
 	VMOVDQU XWORD0, -16(BX)
-	VMOVDQU XWORD4, (SI)
+
 	JMP avxCbcSm4Done
 
 avxCbcSm4Single32:
 	VMOVDQU -32(DX), XWORD0
 	VMOVDQU -16(DX), XWORD1
 
-	VMOVDQU XWORD0, XWORD4
-
 	AVX_SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	VPXOR 0(SI), XWORD0, XWORD0
@@ -396,7 +387,7 @@ avxCbcSm4Single32:
 
 	VMOVDQU XWORD0, -32(BX)
 	VMOVDQU XWORD1, -16(BX)
-	VMOVDQU XWORD4, (SI)
+
 	JMP avxCbcSm4Done
 
 avxCbcSm4Single48:
@@ -404,8 +395,6 @@ avxCbcSm4Single48:
 	VMOVDQU -32(DX), XWORD1
 	VMOVDQU -16(DX), XWORD2
 
-	VMOVDQU XWORD0, XWORD4
-
 	AVX_SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	VPXOR 0(SI), XWORD0, XWORD0
@@ -415,9 +404,9 @@ avxCbcSm4Single48:
 	VMOVDQU XWORD0, -48(BX)
 	VMOVDQU XWORD1, -32(BX)
 	VMOVDQU XWORD2, -16(BX)
-	VMOVDQU XWORD4, (SI)
 	
 avxCbcSm4Done:
+	VMOVDQU X15, (SI)
 	RET
 
 avx2Start:
@@ -425,6 +414,8 @@ avx2Start:
 	VBROADCASTI128 flip_mask<>(SB), BYTE_FLIP_MASK
 	VBROADCASTI128 bswap_mask<>(SB), BSWAP_MASK
 
+	VMOVDQU -16(DX), X15
+
 avx2_16blocks:
 	CMPQ DI, $256
 	JLE avx2CbcSm4Octets
@@ -572,8 +563,6 @@ avx2CbCSm4Single:
 	VMOVDQU -32(DX), XWORD2
 	VMOVDQU -16(DX), XWORD3
 
-	VMOVDQU XWORD0, XWORD4
-
 	AVX_SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	VPXOR 0(SI), XWORD0, XWORD0
@@ -585,20 +574,17 @@ avx2CbCSm4Single:
 	VMOVDQU XWORD1, -48(BX)
 	VMOVDQU XWORD2, -32(BX)
 	VMOVDQU XWORD3, -16(BX)
-	VMOVDQU XWORD4, (SI)
+
 	JMP avx2CbcSm4Done
 
 avx2CbcSm4Single16:
 	VMOVDQU -16(DX), XWORD0
 
-	VMOVDQU XWORD0, XWORD4
-
 	AVX_SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	VPXOR 0(SI), XWORD0, XWORD0
 
 	VMOVDQU XWORD0, -16(BX)
-	VMOVDQU XWORD4, (SI)
 
 	JMP avx2CbcSm4Done
 
@@ -606,8 +592,6 @@ avx2CbcSm4Single32:
 	VMOVDQU -32(DX), XWORD0
 	VMOVDQU -16(DX), XWORD1
 
-	VMOVDQU XWORD0, XWORD4
-
 	AVX_SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	VPXOR 0(SI), XWORD0, XWORD0
@@ -615,7 +599,6 @@ avx2CbcSm4Single32:
 
 	VMOVDQU XWORD0, -32(BX)
 	VMOVDQU XWORD1, -16(BX)
-	VMOVDQU XWORD4, (SI)
 
 	JMP avx2CbcSm4Done
 
@@ -624,8 +607,6 @@ avx2CbcSm4Single48:
 	VMOVDQU -32(DX), XWORD1
 	VMOVDQU -16(DX), XWORD2
 
-	VMOVDQU XWORD0, XWORD4
-
 	AVX_SM4_4BLOCKS(AX, XWORD, YWORD, XWTMP0, XWTMP1, XWORD0, XWORD1, XWORD2, XWORD3)
 
 	VPXOR 0(SI), XWORD0, XWORD0
@@ -635,8 +616,8 @@ avx2CbcSm4Single48:
 	VMOVDQU XWORD0, -48(BX)
 	VMOVDQU XWORD1, -32(BX)
 	VMOVDQU XWORD2, -16(BX)
-	VMOVDQU XWORD4, (SI)
 	
 avx2CbcSm4Done:
+	VMOVDQU X15, (SI)
 	VZEROUPPER
 	RET
diff --git a/sm4/cbc_arm64.s b/sm4/cbc_arm64.s
index 1449e5d..7b42c6d 100644
--- a/sm4/cbc_arm64.s
+++ b/sm4/cbc_arm64.s
@@ -114,6 +114,10 @@ TEXT ·decryptBlocksChain(SB),NOSPLIT,$0
 	MOVD rk, rkSave
 	VLD1 (R6), [IV.B16]
 
+	ADD srcPtr, srcPtrLen, R10
+	SUB $16, R10, R10
+	VLD1 (R10), [V15.S4]
+
 cbcSm4Octets:
 	CMP	$128, srcPtrLen
 	BLE	cbcSm4Nibbles
@@ -233,7 +237,6 @@ cbcSm4Single:
 
 	// 4 blocks
 	VLD1 (srcPtr), [t0.S4, t1.S4, t2.S4, t3.S4]
-	VMOV t0.B16, t4.B16
 	VREV32 t0.B16, t0.B16
 	VREV32 t1.B16, t1.B16
 	VREV32 t2.B16, t2.B16
@@ -263,13 +266,11 @@ cbc4BlocksLoop64:
 	VEOR V8.B16, t3.B16, t3.B16
 
 	VST1 [t0.S4, t1.S4, t2.S4, t3.S4], (dstPtr)
-	VST1 [t4.S4], (R6)
 
 	B cbcSm4Done
 
 cbcSm4Single16:
 	VLD1 (srcPtr), [t0.S4]
-	VMOV t0.B16, t4.B16
 	VREV32 t0.B16, t0.B16
 	VMOV t0.S[1], t1.S[0]
 	VMOV t0.S[2], t2.S[0]
@@ -293,13 +294,11 @@ cbc4BlocksLoop16:
 	VEOR IV.B16, t3.B16, t3.B16
 
 	VST1 [t3.S4], (dstPtr)
-	VST1 [t4.S4], (R6)
 
 	B cbcSm4Done
 
 cbcSm4Single32:
 	VLD1 (srcPtr), [t0.S4, t1.S4]
-	VMOV t0.B16, t4.B16
 	VREV32 t0.B16, t0.B16
 	VREV32 t1.B16, t1.B16
 	PRE_TRANSPOSE_MATRIX(t0, t1, t2, t3, x, y, XTMP6, XTMP7)
@@ -323,12 +322,10 @@ cbc4BlocksLoop32:
 	VEOR V6.B16, t1.B16, t1.B16
 
 	VST1 [t0.S4, t1.S4], (dstPtr)
-	VST1 [t4.S4], (R6)
 	B cbcSm4Done
 
 cbcSm4Single48:
 	VLD1 (srcPtr), [t0.S4, t1.S4, t2.S4]
-	VMOV t0.B16, t4.B16
 	VREV32 t0.B16, t0.B16
 	VREV32 t1.B16, t1.B16
 	VREV32 t2.B16, t2.B16
@@ -355,7 +352,7 @@ cbc4BlocksLoop48:
 	VEOR V7.B16, t2.B16, t2.B16
 
 	VST1 [t0.S4, t1.S4, t2.S4], (dstPtr)
-	VST1 [t4.S4], (R6)
 
 cbcSm4Done:
+	VST1 [V15.S4], (R6)
 	RET