diff --git a/src/internal/bytealg/compare_arm64.s b/src/internal/bytealg/compare_arm64.s
index 9b6354715a..2bd38064c3 100644
--- a/src/internal/bytealg/compare_arm64.s
+++ b/src/internal/bytealg/compare_arm64.s
@@ -10,7 +10,7 @@ TEXT ·Compare(SB),NOSPLIT|NOFRAME,$0-56
 	MOVD	a_len+8(FP), R0
 	MOVD	b_base+24(FP), R3
 	MOVD	b_len+32(FP), R1
-	ADD	$56, RSP, R7
+	MOVD	$ret+48(FP), R7
 	B	cmpbody<>(SB)
 
 TEXT bytes·Compare(SB),NOSPLIT|NOFRAME,$0-56
@@ -18,7 +18,7 @@ TEXT bytes·Compare(SB),NOSPLIT|NOFRAME,$0-56
 	MOVD	a_len+8(FP), R0
 	MOVD	b_base+24(FP), R3
 	MOVD	b_len+32(FP), R1
-	ADD	$56, RSP, R7
+	MOVD	$ret+48(FP), R7
 	B	cmpbody<>(SB)
 
 TEXT runtime·cmpstring(SB),NOSPLIT|NOFRAME,$0-40
@@ -26,7 +26,7 @@ TEXT runtime·cmpstring(SB),NOSPLIT|NOFRAME,$0-40
 	MOVD	a_len+8(FP), R0
 	MOVD	b_base+16(FP), R3
 	MOVD	b_len+24(FP), R1
-	ADD	$40, RSP, R7
+	MOVD	$ret+32(FP), R7
 	B	cmpbody<>(SB)
 
 // On entry:
@@ -37,30 +37,98 @@ TEXT runtime·cmpstring(SB),NOSPLIT|NOFRAME,$0-40
 // R7 points to return value (-1/0/1 will be written here)
 //
 // On exit:
-// R4, R5, and R6 are clobbered
+// R4, R5, R6, R8, R9 and R10 are clobbered
 TEXT cmpbody<>(SB),NOSPLIT|NOFRAME,$0-0
 	CMP	R2, R3
-	BEQ	samebytes // same starting pointers; compare lengths
+	BEQ	samebytes         // same starting pointers; compare lengths
 	CMP	R0, R1
-	CSEL    LT, R1, R0, R6 // R6 is min(R0, R1)
+	CSEL	LT, R1, R0, R6    // R6 is min(R0, R1)
 
-	ADD	R2, R6	// R2 is current byte in a, R6 is last byte in a to compare
-loop:
-	CMP	R2, R6
-	BEQ	samebytes // all compared bytes were the same; compare lengths
-	MOVBU.P	1(R2), R4
-	MOVBU.P	1(R3), R5
+	CMP	$0, R6
+	BEQ	samebytes
+	BIC	$0xf, R6, R10
+	CBZ	R10, small        // length < 16
+	ADD	R2, R10           // end of chunk16
+	// length >= 16
+chunk16_loop:
+	LDP.P	16(R2), (R4, R8)
+	LDP.P	16(R3), (R5, R9)
 	CMP	R4, R5
-	BEQ	loop
-	// bytes differed
+	BNE	cmp
+	CMP	R8, R9
+	BNE	cmpnext
+	CMP	R10, R2
+	BNE	chunk16_loop
+	AND	$0xf, R6, R6
+	CBZ	R6, samebytes
+	SUBS	$8, R6
+	BLT	tail
+	// the length of tail > 8 bytes
+	MOVD.P	8(R2), R4
+	MOVD.P	8(R3), R5
+	CMP	R4, R5
+	BNE	cmp
+	SUB	$8, R6
+	// compare last 8 bytes
+tail:
+	MOVD	(R2)(R6), R4
+	MOVD	(R3)(R6), R5
+	CMP	R4, R5
+	BEQ	samebytes
+cmp:
+	REV	R4, R4
+	REV	R5, R5
+	CMP	R4, R5
+ret:
 	MOVD	$1, R4
-	CSNEG	LT, R4, R4, R4
+	CNEG	HI, R4, R4
 	MOVD	R4, (R7)
 	RET
+small:
+	TBZ	$3, R6, lt_8
+	MOVD	(R2), R4
+	MOVD	(R3), R5
+	CMP	R4, R5
+	BNE	cmp
+	SUBS	$8, R6
+	BEQ	samebytes
+	ADD	$8, R2
+	ADD	$8, R3
+	SUB	$8, R6
+	B	tail
+lt_8:
+	TBZ	$2, R6, lt_4
+	MOVWU	(R2), R4
+	MOVWU	(R3), R5
+	CMPW	R4, R5
+	BNE	cmp
+	SUBS	$4, R6
+	BEQ	samebytes
+	ADD	$4, R2
+	ADD	$4, R3
+lt_4:
+	TBZ	$1, R6, lt_2
+	MOVHU	(R2), R4
+	MOVHU	(R3), R5
+	CMPW	R4, R5
+	BNE	cmp
+	ADD	$2, R2
+	ADD	$2, R3
+lt_2:
+	TBZ	$0, R6, samebytes
+one:
+	MOVBU	(R2), R4
+	MOVBU	(R3), R5
+	CMPW	R4, R5
+	BNE	ret
 samebytes:
-	MOVD	$1, R4
-	CMP	R0, R1
-	CSNEG	LT, R4, R4, R4
-	CSEL	EQ, ZR, R4, R4
+	CMP	R1, R0
+	CSET	NE, R4
+	CNEG	LO, R4, R4
 	MOVD	R4, (R7)
 	RET
+cmpnext:
+	REV	R8, R4
+	REV	R9, R5
+	CMP	R4, R5
+	B	ret