diff --git a/sm4/cbc_amd64.s b/sm4/cbc_amd64.s
index 8d35e57..5895413 100644
--- a/sm4/cbc_amd64.s
+++ b/sm4/cbc_amd64.s
@@ -211,6 +211,8 @@ 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
@@ -222,22 +224,28 @@ 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
@@ -245,6 +253,7 @@ cbcSm4Single32:
 
 	MOVUPS XWORD0, -32(BX)
 	MOVUPS XWORD1, -16(BX)
+	MOVUPS XWORD4, (SI)
 	JMP cbcSm4Done
 
 cbcSm4Single48:
@@ -252,6 +261,8 @@ 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
@@ -261,6 +272,7 @@ cbcSm4Single48:
 	MOVUPS XWORD0, -48(BX)
 	MOVUPS XWORD1, -32(BX)
 	MOVUPS XWORD2, -16(BX)
+	MOVUPS XWORD4, (SI)
 	
 cbcSm4Done:
 	RET
@@ -342,6 +354,8 @@ 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
@@ -353,22 +367,28 @@ 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
@@ -376,6 +396,7 @@ avxCbcSm4Single32:
 
 	VMOVDQU XWORD0, -32(BX)
 	VMOVDQU XWORD1, -16(BX)
+	VMOVDQU XWORD4, (SI)
 	JMP avxCbcSm4Done
 
 avxCbcSm4Single48:
@@ -383,6 +404,8 @@ 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
@@ -392,6 +415,7 @@ avxCbcSm4Single48:
 	VMOVDQU XWORD0, -48(BX)
 	VMOVDQU XWORD1, -32(BX)
 	VMOVDQU XWORD2, -16(BX)
+	VMOVDQU XWORD4, (SI)
 	
 avxCbcSm4Done:
 	RET
@@ -548,6 +572,8 @@ 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
@@ -559,22 +585,29 @@ 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
 
 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
@@ -582,6 +615,8 @@ avx2CbcSm4Single32:
 
 	VMOVDQU XWORD0, -32(BX)
 	VMOVDQU XWORD1, -16(BX)
+	VMOVDQU XWORD4, (SI)
+
 	JMP avx2CbcSm4Done
 
 avx2CbcSm4Single48:
@@ -589,6 +624,8 @@ 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
@@ -598,6 +635,7 @@ avx2CbcSm4Single48:
 	VMOVDQU XWORD0, -48(BX)
 	VMOVDQU XWORD1, -32(BX)
 	VMOVDQU XWORD2, -16(BX)
+	VMOVDQU XWORD4, (SI)
 	
 avx2CbcSm4Done:
 	VZEROUPPER
diff --git a/sm4/cbc_arm64.s b/sm4/cbc_arm64.s
index a9c978f..1449e5d 100644
--- a/sm4/cbc_arm64.s
+++ b/sm4/cbc_arm64.s
@@ -233,6 +233,7 @@ 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
@@ -262,11 +263,13 @@ 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]
@@ -290,11 +293,13 @@ 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)
@@ -318,10 +323,12 @@ 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
@@ -348,6 +355,7 @@ cbc4BlocksLoop48:
 	VEOR V7.B16, t2.B16, t2.B16
 
 	VST1 [t0.S4, t1.S4, t2.S4], (dstPtr)
+	VST1 [t4.S4], (R6)
 
 cbcSm4Done:
 	RET
diff --git a/sm4/cbc_cipher_asm.go b/sm4/cbc_cipher_asm.go
index 423fa05..582b051 100644
--- a/sm4/cbc_cipher_asm.go
+++ b/sm4/cbc_cipher_asm.go
@@ -19,7 +19,6 @@ const cbcDecrypt = 0
 type cbc struct {
 	b   *sm4CipherAsm
 	iv  []byte
-	tmp []byte
 	enc int
 }
 
@@ -28,7 +27,6 @@ func (b *sm4CipherAsm) NewCBCEncrypter(iv []byte) cipher.BlockMode {
 	c.b = b
 	c.enc = cbcEncrypt
 	c.iv = make([]byte, BlockSize)
-	c.tmp = make([]byte, BlockSize)
 	copy(c.iv, iv)
 	return &c
 }
@@ -38,7 +36,6 @@ func (b *sm4CipherAsm) NewCBCDecrypter(iv []byte) cipher.BlockMode {
 	c.b = b
 	c.enc = cbcDecrypt
 	c.iv = make([]byte, BlockSize)
-	c.tmp = make([]byte, BlockSize)
 	copy(c.iv, iv)
 	return &c
 }
@@ -68,16 +65,8 @@ func (x *cbc) CryptBlocks(dst, src []byte) {
 		encryptBlocksChain(&x.b.enc[0], dst, src, &x.iv[0])
 		return
 	}
-	// For each block, we need to xor the decrypted data with the previous block's ciphertext (the iv).
-	// To avoid making a copy each time, we loop over the blocks BACKWARDS.
-	end := len(src)
-	// Copy the last block of ciphertext in preparation as the new iv.
-	copy(x.tmp, src[end-BlockSize:end])
 
 	decryptBlocksChain(&x.b.dec[0], dst, src, &x.iv[0])
-
-	// Set the new iv to the first block we copied earlier.
-	x.iv, x.tmp = x.tmp, x.iv
 }
 
 func (x *cbc) SetIV(iv []byte) {