// result: (DUFFCOPY [16 * (128 - s/8)] dst src mem)
	for {
		s := auxIntToInt64(v.AuxInt)
		t := auxToType(v.Aux)
		dst := v_0
		src := v_1
		mem := v_2
		if !(s%8 == 0 && s >= 24 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice && logLargeCopy(v, s)) {
			break
		}
		v.reset(OpLOONG64DUFFCOPY)

		(OffPtr <src.Type> src [s-16])
		(Move [s-s%16] dst src mem))

// medium move uses a duff device
(Move [s] dst src mem)
	&& s > 64 && s <= 16*64 && s%16 == 0
	&& !config.noDuffDevice && logLargeCopy(v, s) =>
	(DUFFCOPY [8 * (64 - s/16)] dst src mem)
// 8 is the number of bytes to encode:
//
// LDP.P   16(R16), (R26, R27)
// STP.P   (R26, R27), 16(R17)
//

	// result: (DUFFCOPY [16 * (128 - s/8)] dst src mem)
	for {
		s := auxIntToInt64(v.AuxInt)
		t := auxToType(v.Aux)
		dst := v_0
		src := v_1
		mem := v_2
		if !(s%8 == 0 && s >= 24 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice && logLargeCopy(v, s)) {
			break
		}
		v.reset(OpMIPS64DUFFCOPY)

	// cond: s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice && logLargeCopy(v, s)
	// result: (DUFFCOPY [16 * (128 - s/8)] dst src mem)
	for {
		s := auxIntToInt64(v.AuxInt)
		t := auxToType(v.Aux)
		dst := v_0
		src := v_1
		mem := v_2
		if !(s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice && logLargeCopy(v, s)) {
			break
		}
		v.reset(OpRISCV64DUFFCOPY)