MOVH	R7, 4(R6)
	MOVB	R8, 6(R6)
	RET

move8to11:
	CMPBNE	R5, $8, move9
	MOVD	0(R4), R3
	MOVD	R3, 0(R6)
	RET
move9:
	CMPBNE	R5, $9, move10
	MOVD	0(R4), R3
	MOVB	8(R4), R7
	MOVD	R3, 0(R6)
	MOVB	R7, 8(R6)
	RET
move10:
	CMPBNE	R5, $10, move11
	MOVD	0(R4), R3
	MOVH	8(R4), R7
	MOVD	R3, 0(R6)
	MOVH	R7, 8(R6)
	RET
move11:
	MOVD	0(R4), R3
	MOVH	8(R4), R7

	ADDU	R3, R8	// R3 is current byte in a, R8 is last byte in a to compare
loop:
	BEQ	R3, R8, samebytes

	MOVBU	(R3), R6
	ADDU	$1, R3
	MOVBU	(R4), R7
	ADDU	$1, R4
	BEQ	R6, R7 , loop

	SGTU	R6, R7, R8
	MOVW	$-1, R6
	CMOVZ	R8, R6, R8
	JMP	cmp_ret
samebytes:
	SGTU	R1, R2, R6
	SGTU	R2, R1, R7
	SUBU	R7, R6, R8
cmp_ret:
	MOVW	R8, ret+24(FP)
	RET

TEXT runtime·cmpstring(SB),NOSPLIT,$0-20
	MOVW	a_base+0(FP), R3

	// make sure both a and b are aligned.
	OR	R4, R6, R15
	AND	$7, R15
	BNE	R0, R15, byte_loop

	PCALIGN	$16
chunk16_loop:
	BEQ	R0, R14, byte_loop
	MOVV	(R4), R8
	MOVV	(R6), R9
	BNE	R8, R9, byte_loop
	MOVV	8(R4), R16
	MOVV	8(R6), R17
	ADDV	$16, R4
	ADDV	$16, R6
	SUBVU	$1, R14
	BEQ	R16, R17, chunk16_loop
	SUBV	$8, R4
	SUBV	$8, R6

byte_loop:
	BEQ	R4, R12, samebytes
	MOVBU	(R4), R8

	// incoming:
	// R3 a addr -> R5
	// R4 a len  -> R3
	// R5 b addr -> R6
	// R6 b len  -> R4
	//
	// on entry to cmpbody:
	// R3 compare value if compared length is same.
	// R5 a addr
	// R6 b addr
	// R9 min(len(a),len(b))
	SETB_INIT()
	CMP	R4,R6,CR0
	CMP	R3,R5,CR7
	ISEL	CR0LT,R4,R6,R9
	MOVD	R5,R6
	MOVD	R3,R5
	SETB_CR0(R3)
	BC	$12,30,LR	// beqlr cr7
	BR	cmpbody<>(SB)

	SGTU	R1, R2, R6
	BNE	R6, backward

	// if less than 4 bytes, use byte by byte copying
	SGTU	$4, R3, R6
	BNE	R6, f_small_copy

	// align destination to 4 bytes
	AND	$3, R1, R6
	BEQ	R6, f_dest_aligned
	SUBU	R1, R0, R6
	AND	$3, R6
	MOVWHI	0(R2), R7
	SUBU	R6, R3
	MOVWLO	3(R2), R7
	ADDU	R6, R2
	MOVWHI	R7, 0(R1)
	ADDU	R6, R1

f_dest_aligned:
	AND	$31, R3, R7

	CMP	R4, R5
	BNE	cmp
	CMP	R8, R9
	BNE	cmpnext
	CMP	R10, R0
	BNE	chunk16_loop
	AND	$0xf, R6, R6
	CBZ	R6, samebytes
	SUBS	$8, R6
	BLT	tail
	// the length of tail > 8 bytes
	MOVD.P	8(R0), R4
	MOVD.P	8(R2), R5
	CMP	R4, R5
	BNE	cmp
	SUB	$8, R6
	// compare last 8 bytes
tail:
	MOVD	(R0)(R6), R4
	MOVD	(R2)(R6), R5
	CMP	R4, R5
	BEQ	samebytes
cmp:
	REV	R4, R4
	REV	R5, R5
	CMP	R4, R5
ret:

A6:
	MOVD   (R8)(R1*1), R6
	MULHDU R9, R6
	MOVD   (R2)(R1*1), R10
	ADDC   R10, R11        // add to low order bits
	ADDE   R0, R6
	ADDC   R4, R11
	ADDE   R0, R6
	MOVD   R6, R4
	MOVD   R11, (R2)(R1*1)

	MOVD   (8)(R8)(R1*1), R6
	MULHDU R9, R6
	MOVD   (8)(R2)(R1*1), R10
	ADDC   R10, R11           // add to low order bits
	ADDE   R0, R6
	ADDC   R4, R11
	ADDE   R0, R6
	MOVD   R6, R4

	SUB	R3, R4
	MOVW	$0, R7

	MOVW.W	-4(R2), R6
	MOVW	R6<<R3, R7
	MOVW	R6>>R4, R6
	MOVW	R6, c+28(FP)
	B E7

L7:
	MOVW.W	-4(R2), R6
	ORR	R6>>R4, R7
	MOVW.W	R7, -4(R5)
	MOVW	R6<<R3, R7
E7:
	TEQ	R1, R5
	BNE	L7

	MOVW	R7, -4(R5)
	RET

Y7:	// copy loop, because shift 0 == shift 32
	MOVW.W	-4(R2), R6
	MOVW.W	R6, -4(R5)
	TEQ	R1, R5
	BNE Y7

X7:

	ROUND3(R4,R5,R6,R7, 8,0xfffa3942, 4);
	ROUND3(R7,R4,R5,R6,11,0x8771f681,11);
	ROUND3(R6,R7,R4,R5,14,0x6d9d6122,16);
	ROUND3(R5,R6,R7,R4, 1,0xfde5380c,23);
	ROUND3(R4,R5,R6,R7, 4,0xa4beea44, 4);
	ROUND3(R7,R4,R5,R6, 7,0x4bdecfa9,11);
	ROUND3(R6,R7,R4,R5,10,0xf6bb4b60,16);
	ROUND3(R5,R6,R7,R4,13,0xbebfbc70,23);
	ROUND3(R4,R5,R6,R7, 0,0x289b7ec6, 4);
	ROUND3(R7,R4,R5,R6, 3,0xeaa127fa,11);
	ROUND3(R6,R7,R4,R5, 6,0xd4ef3085,16);

	MULLU	R1,R3,(R6,R5)		// R7:R6:R5 = R2:R1 * R3
	MOVW	$0,R7
	MULALU	R2,R3,(R7,R6)

	// unscale by discarding low 32 bits, shifting the rest by 29
	MOVW	R6>>29,R6		// R7:R6 = (R7:R6:R5 >> 61)
	ORR	R7<<3,R6
	MOVW	R7>>29,R7

	// subtract (10**9 * sec) from nsec to get nanosecond remainder
	MOVW	$1000000000, R5	// 10**9
	MULLU	R6,R5,(R9,R8)   // R9:R8 = R7:R6 * R5
	MULA	R7,R5,R9,R9