instr	$0, R5, R9;		\
	instr	$0, R6, R10;		\
	instr	$0, R7, R11;		\
	STP.P	(R8, R9), 32(R3);	\
	STP	(R10, R11), -16(R3);	\
	SUB	$4, counter;

// do one iteration of copy in addVW/subVW
#define vwOneIterCopy(counter, exit)			\
	CBZ	counter, exit;				\
	LDP.P	32(R1), (R4, R5);			\
	LDP	-16(R1), (R6, R7);			\
	STP.P	(R4, R5), 32(R3);			\
	STP	(R6, R7), -16(R3);			\
	SUB	$4, counter;

	ADD R0>>51, R11, R11
	ADD R1>>51, R12, R12
	ADD R2>>51, R13, R13
	ADD R3>>51, R14, R14
	// R4>>51 * 19 + R10 -> R10
	LSR $51, R4, R21
	MOVD $19, R22
	MADD R22, R10, R21, R10

	STP (R10, R11), 0(R20)
	STP (R12, R13), 16(R20)
	MOVD R14, 32(R20)

EXT(crosscall1):
	.cfi_startproc
	stp x29, x30, [sp, #-96]!
	.cfi_def_cfa_offset 96
	.cfi_offset 29, -96
	.cfi_offset 30, -88
	mov x29, sp
	.cfi_def_cfa_register 29
	stp x19, x20, [sp, #80]
	.cfi_offset 19, -16
	.cfi_offset 20, -8
	stp x21, x22, [sp, #64]
	.cfi_offset 21, -32
	.cfi_offset 22, -24
	stp x23, x24, [sp, #48]
	.cfi_offset 23, -48
	.cfi_offset 24, -40
	stp x25, x26, [sp, #32]
	.cfi_offset 25, -64

		LDP.P	16(b_ptr), (acc2, acc3)
		CSEL	EQ, acc2, t2, t2
		CSEL	EQ, acc3, t3, t3

		CMP	$16, const1
		BNE	loop_select

	STP	(x0, x1), 0*16(res_ptr)
	STP	(x2, x3), 1*16(res_ptr)
	STP	(y0, y1), 2*16(res_ptr)
	STP	(y2, y3), 3*16(res_ptr)
	STP	(t0, t1), 4*16(res_ptr)
	STP	(t2, t3), 5*16(res_ptr)
	RET
/* ---------------------------------------*/

	// but we are not called from Go so that space is ours to use,
	// and we must to be contiguous with the stack arguments.
	MOVD	$arg0-(7*8)(SP), R14
	STP	(R0, R1), (0*8)(R14)
	STP	(R2, R3), (2*8)(R14)
	STP	(R4, R5), (4*8)(R14)
	STP	(R6, R7), (6*8)(R14)

	// Push C callee-save registers R19-R28.
	// LR, FP already saved.
	SAVE_R19_TO_R28(8*9)

	// Create a struct callbackArgs on our stack.

	MUL	R8, R3, R16
	ADCS	R16, R12
	UMULH	R8, R3, R20
	ADC	$0, R20

	MUL	R5, R3, R13
	ADDS	R13, R9
	UMULH	R5, R3, R17
	ADCS	R17, R10
	UMULH	R6, R3, R21
	STP.P	(R9, R10), 16(R1)
	ADCS	R21, R11
	UMULH	R7, R3, R19
	ADCS	R19, R12
	STP.P	(R11, R12), 16(R1)
	ADC	$0, R20, R4

	SUB	$4, R0
	B	loop

done:
	MOVD	R4, c+24(FP)

	// R0 and R1 already saved
	STP	(R2, R3), 1*8(RSP)
	STP	(R4, R5), 3*8(RSP)
	STP	(R6, R7), 5*8(RSP)
	STP	(R8, R9), 7*8(RSP)
	STP	(R10, R11), 9*8(RSP)
	STP	(R12, R13), 11*8(RSP)
	STP	(R14, R15), 13*8(RSP)
	// R16, R17 may be clobbered by linker trampoline
	// R18 is unused.
	STP	(R19, R20), 15*8(RSP)
	STP	(R21, R22), 17*8(RSP)
	STP	(R23, R24), 19*8(RSP)
	STP	(R25, R26), 21*8(RSP)

	 *  push 3 args for fn (R0, R1, R3), skipping R2.
	 * Also note that at procedure entry in gc world, 8(RSP) will be the
	 *  first arg.
	 */
	SUB	$(8*24), RSP
	STP	(R0, R1), (8*1)(RSP)
	MOVD	R3, (8*3)(RSP)

	SAVE_R19_TO_R28(8*4)
	SAVE_F8_TO_F15(8*14)
	STP	(R29, R30), (8*22)(RSP)


	// Initialize Go ABI environment
	BL	runtime·load_g(SB)
	BL	runtime·cgocallback(SB)

	RESTORE_R19_TO_R28(8*4)

	LDPSW	8(R1), (ZR, R2)     // 3f084169
	STP	(R3, R4), (R5)      // a31000a9
	STP	(R3, R4), 8(R5)     // a39000a9
	STP.W	(R3, R4), 8(R5)     // a39080a9
	STP.P	(R3, R4), 8(R5)     // a39080a8
	STP	(R3, R4), -8(R5)    // a3903fa9
	STP	(R3, R4), -4(R5)    // bb1000d1631300a9
	STP	(R3, R4), 11(R0)    // 1b2c0091631300a9
	STP	(R3, R4), 1024(R0)  // 1b001091631300a9
	STP	(R3, R4), (RSP)     // e31300a9

	_, _ = w1[3], w2[3]
	w1[0], w1[1], w1[2], w1[3] = 0, 0, 0, 0 // arm64:"STP",-"MOVB",-"MOVH"
	w2[2], w2[3], w2[1], w2[0] = 0, 0, 0, 0 // arm64:"STP",-"MOVB",-"MOVH"
}

func zero_uint64_2(d1, d2 []uint64) {
	_, _ = d1[1], d2[1]
	d1[0], d1[1] = 0, 0 // arm64:"STP",-"MOVB",-"MOVH"
	d2[1], d2[0] = 0, 0 // arm64:"STP",-"MOVB",-"MOVH"
}

func loadstore(p, q *[4]uint8) {
	// amd64:"MOVL",-"MOVB"