MOVW	x+12(FP), R2
	MOVW	y+24(FP), R3
	ADD	R5<<2, R1, R5
	MOVW	$0, R4
	B E9

	// word loop
L9:
	MOVW.P	4(R2), R6
	MULLU	R6, R3, (R7, R6)
	ADD.S	R4, R6
	ADC	R0, R7
	MOVW	0(R1), R4
	ADD.S	R4, R6
	ADC	R0, R7
	MOVW.P	R6, 4(R1)
	MOVW	R7, R4
E9:
	TEQ	R1, R5
	BNE	L9

	MOVW	R4, c+28(FP)

	MOVW	c, R9; \
	ADDW	b, a

	ROUND1(R4,R5,R6,R7, 1,0xd76aa478, 7);
	ROUND1(R7,R4,R5,R6, 2,0xe8c7b756,12);
	ROUND1(R6,R7,R4,R5, 3,0x242070db,17);
	ROUND1(R5,R6,R7,R4, 4,0xc1bdceee,22);
	ROUND1(R4,R5,R6,R7, 5,0xf57c0faf, 7);
	ROUND1(R7,R4,R5,R6, 6,0x4787c62a,12);
	ROUND1(R6,R7,R4,R5, 7,0xa8304613,17);
	ROUND1(R5,R6,R7,R4, 8,0xfd469501,22);
	ROUND1(R4,R5,R6,R7, 9,0x698098d8, 7);
	ROUND1(R7,R4,R5,R6,10,0x8b44f7af,12);

	ROUND3(R2,R3,R4,R5, 8,0xfffa3942, 4);
	ROUND3(R5,R2,R3,R4,11,0x8771f681,11);
	ROUND3(R4,R5,R2,R3,14,0x6d9d6122,16);
	ROUND3(R3,R4,R5,R2, 1,0xfde5380c,23);
	ROUND3(R2,R3,R4,R5, 4,0xa4beea44, 4);
	ROUND3(R5,R2,R3,R4, 7,0x4bdecfa9,11);
	ROUND3(R4,R5,R2,R3,10,0xf6bb4b60,16);
	ROUND3(R3,R4,R5,R2,13,0xbebfbc70,23);
	ROUND3(R2,R3,R4,R5, 0,0x289b7ec6, 4);
	ROUND3(R5,R2,R3,R4, 3,0xeaa127fa,11);
	ROUND3(R4,R5,R2,R3, 6,0xd4ef3085,16);

	MOVW	0(R14), R3		// arg 1 - fd
	MOVW	4(R14), R4		// arg 2 - cmd
	MOVW	8(R14), R5		// arg 3 - arg
	MOVD	$0, R6			// vararg
	CALL	libc_fcntl(SB)
	MOVD	$0, R4
	CMP	R3, $-1
	BNE	noerr
	CALL	libc_errno(SB)
	MOVW	(R3), R4		// errno
	MOVW	$-1, R3
noerr:
	MOVW	R3, 12(R14)
	MOVW	R4, 16(R14)
	RET

TEXT runtime·sigaction_trampoline(SB),NOSPLIT,$32
	MOVD	8(R3), R4		// arg 2 - new
	MOVD	16(R3), R5		// arg 3 - old

	ROUND3(R5,R22,R3,R4,M00,0xeaa127fa,11);
	ROUND3(R4,R5,R22,R3,M03,0xd4ef3085,16);
	ROUND3(R3,R4,R5,R22,M06, 0x4881d05,23);
	ROUND3(R22,R3,R4,R5,M09,0xd9d4d039, 4);
	ROUND3(R5,R22,R3,R4,M12,0xe6db99e5,11);
	ROUND3(R4,R5,R22,R3,M15,0x1fa27cf8,16);
	ROUND3(R3,R4,R5,R22,M02,0xc4ac5665,23);

	ROUND4(R22,R3,R4,R5,M00,0xf4292244, 6);
	ROUND4(R5,R22,R3,R4,M07,0x432aff97,10);
	ROUND4(R4,R5,R22,R3,M14,0xab9423a7,15);

	BDNZ	loop64

	ISEL	CR0EQ, R11, R3, R3	// If no tail, return 1, otherwise R3 remains 0.
	BEQLR				// return if no tail.

	ADD	$-64, R9, R8
	ADD	$-64, R10, R4
	LXVD2X	(R8+R0), V0
	LXVD2X	(R4+R0), V1
	VCMPEQUBCC	V0, V1, V2
	BGELR_CR6
	LXVD2X	(R8+R14), V0
	LXVD2X	(R4+R14), V1
	VCMPEQUBCC	V0, V1, V2
	BGELR_CR6
	LXVD2X	(R8+R15), V0
	LXVD2X	(R4+R15), V1
	VCMPEQUBCC	V0, V1, V2
	BGELR_CR6
	LXVD2X	(R8+R16), V0

	// make sure both a and b are aligned.
	OR	R3, R4, R11
	AND	$7, R11
	BNE	R0, R11, byte_loop

chunk16_loop:
	BEQ	R0, R10, byte_loop
	MOVV	(R3), R6
	MOVV	(R4), R7
	BNE	R6, R7, byte_loop
	MOVV	8(R3), R13
	MOVV	8(R4), R14
	ADDV	$16, R3
	ADDV	$16, R4
	SUBVU	$1, R10
	BEQ	R13, R14, chunk16_loop
	SUBV	$8, R3
	SUBV	$8, R4

byte_loop:
	BEQ	R3, R8, samebytes
	MOVBU	(R3), R6

	CBZ	R0, len0	// the length of z is 0
	MOVD.P	8(R1), R4
	ADDS	R2, R4		// z[0] = x[0] + y, set carry
	MOVD.P	R4, 8(R3)
	SUB	$1, R0
	CBZ	R0, len1	// the length of z is 1
	TBZ	$0, R0, two
	MOVD.P	8(R1), R4	// do it once
	ADCS	$0, R4
	MOVD.P	R4, 8(R3)
	SUB	$1, R0
two:				// do it twice
	TBZ	$1, R0, loop
	LDP.P	16(R1), (R4, R5)
	ADCS	$0, R4, R8	// c, z[i] = x[i] + c
	ADCS	$0, R5, R9

	DMB	MB_ISH
	MOVW	time_hi2(R3), R2
	CMP	R1, R2
	BNE	loop

	// wintime = R1:R0, multiply by 100
	MOVW	$100, R2
	MULLU	R0, R2, (R4, R3)    // R4:R3 = R1:R0 * R2
	MULA	R1, R2, R4, R4

	// wintime*100 = R4:R3
	MOVW	R3, ret_lo+0(FP)
	MOVW	R4, ret_hi+4(FP)
	RET

// save_g saves the g register (R10) into thread local memory
// so that we can call externally compiled

	BNE	repeat
	MOVD	R6, ret+16(FP)
	RET

// func Xchg(ptr *uint32, new uint32) uint32
TEXT ·Xchg(SB), NOSPLIT, $0-20
	MOVD	ptr+0(FP), R4
	MOVW	new+8(FP), R3
	MOVW	(R4), R6
repeat:
	CS	R6, R3, (R4) // if R6==(R4) then (R4)=R3 else R6=(R4)
	BNE	repeat
	MOVW	R6, ret+16(FP)
	RET

// func Xchg64(ptr *uint64, new uint64) uint64
TEXT ·Xchg64(SB), NOSPLIT, $0-24
	MOVD	ptr+0(FP), R4