(Move [s] dst src mem) && s > 0 && s <= 256 && logLargeCopy(v, s) =>
	(MVC [makeValAndOff(int32(s), 0)] dst src mem)
(Move [s] dst src mem) && s > 256 && s <= 512 && logLargeCopy(v, s) =>
	(MVC [makeValAndOff(int32(s)-256, 256)] dst src (MVC [makeValAndOff(256, 0)] dst src mem))
(Move [s] dst src mem) && s > 512 && s <= 768 && logLargeCopy(v, s) =>

(Move [s] dst src mem) && s > 8 && s < 16 =>
	(I64Store [s-8] dst (I64Load [s-8] src mem)
		(I64Store dst (I64Load src mem) mem))

// Large copying uses helper.
(Move [s] dst src mem) && logLargeCopy(v, s) =>
	(LoweredMove [s] dst src mem)

// Lowering Zero instructions
(Zero [0] _ mem) => mem
(Zero [1] destptr mem) => (I64Store8 destptr (I64Const [0]) mem)

			(MOVVstore dst (MOVVload src mem) mem)))

// medium move uses a duff device
(Move [s] {t} dst src mem)
	&& s%8 == 0 && s >= 24 && s <= 8*128 && t.Alignment()%8 == 0
	&& !config.noDuffDevice && logLargeCopy(v, s)  =>
	(DUFFCOPY [16 * (128 - s/8)] dst src mem)
// 16 and 128 are magic constants.  16 is the number of bytes to encode:
//	MOVV	(R1), R23
//	ADDV	$8, R1
//	MOVV	R23, (R2)
//	ADDV	$8, R2

(Move [s] {t} dst src mem)
	&& s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0
	&& !config.noDuffDevice && logLargeCopy(v, s) =>
	(DUFFCOPY [16 * (128 - s/8)] dst src mem)

// Generic move uses a loop
(Move [s] {t} dst src mem) && (s <= 16 || logLargeCopy(v, s)) =>
	(LoweredMove [t.Alignment()]
		dst
		src
		(ADDI <src.Type> [s-moveSize(t.Alignment(), config)] src)
		mem)

// medium move uses a duff device
(Move [s] {t} dst src mem)
	&& s%8 == 0 && s >= 24 && s <= 8*128 && t.Alignment()%8 == 0
	&& !config.noDuffDevice && logLargeCopy(v, s)  =>
	(DUFFCOPY [16 * (128 - s/8)] dst src mem)
// 16 and 128 are magic constants.  16 is the number of bytes to encode:
//	MOVV	(R1), R23
//	ADDV	$8, R1
//	MOVV	R23, (R2)
//	ADDV	$8, R2

	return isInlinableMemmove(dst, src, sz, c)
}

// logLargeCopy logs the occurrence of a large copy.
// The best place to do this is in the rewrite rules where the size of the move is easy to find.
// "Large" is arbitrarily chosen to be 128 bytes; this may change.
func logLargeCopy(v *Value, s int64) bool {
	if s < 128 {
		return true
	}
	if logopt.Enabled() {

// offset is zero, any alignment can be used.
(Move [s] dst src mem) && s > 8 && buildcfg.GOPPC64 <= 8 && logLargeCopy(v, s) =>
        (LoweredMove [s] dst src mem)
(Move [s] dst src mem) && s > 8 && s <= 64 && buildcfg.GOPPC64 >= 9 =>
        (LoweredQuadMoveShort [s] dst src mem)
(Move [s] dst src mem) && s > 8 && buildcfg.GOPPC64 >= 9 && logLargeCopy(v, s) =>
        (LoweredQuadMove [s] dst src mem)

// Calls
// Lowering calls

		v.AddArg3(dst, v0, v1)
		return true
	}
	// match: (Move [s] dst src mem)
	// cond: s > 0 && s <= 256 && logLargeCopy(v, s)
	// result: (MVC [makeValAndOff(int32(s), 0)] dst src mem)
	for {
		s := auxIntToInt64(v.AuxInt)
		dst := v_0
		src := v_1
		mem := v_2
		if !(s > 0 && s <= 256 && logLargeCopy(v, s)) {
			break
		}
		v.reset(OpS390XMVC)
		v.AuxInt = valAndOffToAuxInt(makeValAndOff(int32(s), 0))

			(MOVWstore [4] dst (MOVWload [4] src mem)
				(MOVWstore dst (MOVWload src mem) mem))))


// large or unaligned move uses a loop
(Move [s] {t} dst src mem)
	&& (s > 16 && logLargeCopy(v, s) || t.Alignment()%4 != 0) =>
	(LoweredMove [int32(t.Alignment())]
		dst
		src
		(ADDconst <src.Type> src [int32(s-moveSize(t.Alignment(), config))])
		mem)

// calls

	// 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)