// returns memory
		{
			name:      "DUFFCOPY",
			aux:       "Int64",
			argLength: 3,
			reg: regInfo{
				inputs:   []regMask{buildReg("DI"), buildReg("SI")},
				clobbers: buildReg("DI SI X0"), // uses X0 as a temporary
			},
			clobberFlags:   true,
			faultOnNilArg0: true,
			faultOnNilArg1: true,
			unsafePoint:    true, // FP maintenance around DUFFCOPY can be clobbered by interrupts
		},

				}

				if yt.zcase == Zcallduff && ctxt.Arch.Family == sys.AMD64 {
					// Maintain BP around call, since duffcopy/duffzero can't do it
					// (the call jumps into the middle of the function).
					// This makes it possible to see call sites for duffcopy/duffzero in
					// BP-based profiling tools like Linux perf (which is the
					// whole point of maintaining frame pointers in Go).

		(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)
//
// 64 is number of these blocks. See runtime/duff_arm64.s:duffcopy

		v.AddArg3(dst, v0, v1)
		return true
	}
	// match: (Move [s] {t} dst src mem)
	// cond: s%8 == 0 && s >= 24 && 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

		v.AddArg3(dst, v0, v1)
		return true
	}
	// match: (Move [s] {t} dst src mem)
	// cond: s%8 == 0 && s >= 24 && 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

		v.AddArg3(dst, v0, v1)
		return true
	}
	// match: (Move [s] {t} dst src mem)
	// 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