OR	$65536, R1            // c01d00010000
	OR	$-2, R1               // c0a1fffffffeb981001a
	ORW	R1, R2                // 1621
	ORW	R1, R2, R3            // b9f62031
	ORW	$1, R1                // a51b0001
	ORW	$131071, R1           // c01d0001ffff
	ORW	$65536, R1            // a51a0001
	ORW	$-2, R1               // c01dfffffffe
	XOR	R1, R2                // b9820021
	XOR	R1, R2, R3            // b9e72031

	"MOVW",
	"MOVWZ",
	"MOVD",
	"MOVDBR",
	"MOVDEQ",
	"MOVDGE",
	"MOVDGT",
	"MOVDLE",
	"MOVDLT",
	"MOVDNE",
	"LOCR",
	"LOCGR",
	"FLOGR",
	"POPCNT",
	"AND",
	"ANDW",
	"OR",
	"ORW",
	"XOR",
	"XORW",
	"SLW",
	"SLD",
	"SRW",
	"SRAW",
	"SRD",
	"SRAD",
	"RLL",
	"RLLG",
	"RNSBG",
	"RXSBG",
	"ROSBG",
	"RNSBGT",
	"RXSBGT",
	"ROSBGT",
	"RISBG",
	"RISBGN",

	MOVBZ	val+8(FP), R4
	// We don't have atomic operations that work on individual bytes so we
	// need to align addr down to a word boundary and create a mask
	// containing v to AND with the entire word atomically.
	ORW	$~0xff, R4           // R4 = uint32(v) | 0xffffff00
	MOVD	$(3<<3), R5
	RXSBG	$59, $60, $3, R3, R5 // R5 = 24 - ((addr % 4) * 8) = ((addr & 3) << 3) ^ (3 << 3)
	ANDW	$~3, R3              // R3 = floor(addr, 4) = addr &^ 3

	// ppc64:`MOVHZ`,-`MOVBZ`
	// s390x:`MOVHZ\s\(.*\),`,-`OR`,-`ORW`,-`SLD`,-`SLW`
	return binary.BigEndian.Uint16(b)
}

func load_be16_idx(b []byte, idx int) uint16 {
	// amd64:`ROLW\s\$8`,-`MOVB`,-`OR`
	// arm64:`REV16W`,`MOVHU\s\(R[0-9]+\)\(R[0-9]+\),`,-`MOVB`
	// ppc64le:`MOVHBR`,-`MOVBZ`
	// ppc64:`MOVHZ`,-`MOVBZ`
	// s390x:`MOVHZ\s\(.*\)\(.*\*1\),`,-`OR`,-`ORW`,-`SLD`,-`SLW`
	return binary.BigEndian.Uint16(b[idx:])
}

// (x + y) / 2 with x>=y -> (x - y) / 2 + y
(Avg64u <t> x y) => (ADD (SRDconst <t> (SUB <t> x y) [1]) y)

(And64 ...) => (AND ...)
(And(32|16|8) ...) => (ANDW ...)

(Or64 ...) => (OR ...)
(Or(32|16|8) ...) => (ORW ...)

(Xor64 ...) => (XOR ...)
(Xor(32|16|8) ...) => (XORW ...)

(Neg64 ...) => (NEG ...)
(Neg(32|16|8) ...) => (NEGW ...)
(Neg32F ...) => (FNEGS ...)
(Neg64F ...) => (FNEG ...)

(Com64 ...) => (NOT ...)

			v.AuxInt = int32ToAuxInt(int32(c))
			v.AddArg(x)
			return true
		}
		break
	}
	// match: (ORW x x)
	// result: x
	for {
		x := v_0
		if x != v_1 {
			break
		}
		v.copyOf(x)
		return true
	}
	// match: (ORW <t> x g:(MOVWload [off] {sym} ptr mem))
	// cond: ptr.Op != OpSB && is20Bit(int64(off)) && canMergeLoadClobber(v, g, x) && clobber(g)

rawon??i&dem?mraw?nydg,?k&elo&guld?rtso??slopolam?tsu?ytsyrut??l&ip?o&kzs?w&-awolats?oksnok????n&erapohs,img?zcel,?rog&-ai&bab?nelej??j?z??syn?tsaim?w&a&l&eib?i?o??zsraw??o&namil?tainop,??z&eiwolaib?mol???c&e&iw&alselob?o&nsos?rtso???le&im?zrogz???orw,p??d&em,ia?ragrats?uolc&inu,sds,??e&c&i&lrog?w&ilg,o&hc&arats?orp??klop?tak????yzreibok??i&csjuoniws?ksromop?saldop??l&ahdop?opo??napokaz,t&atselaer?iselpmis,?z&romop?swozam???g&alble?ezrbo&lok?nrat??ro??hcyzrblaw?i&csomohcurein?grat?klawus??k&e&ru...

rawon??i&dem?mraw?nydg,?k&elo&guld?rtso??slopolam?tsu?ytsyrut??l&ip?o&kzs?w&-awolats?oksnok????n&erapohs,img?zcel,?rog&-ai&bab?nelej??j?z??syn?tsaim?w&a&l&eib?i?o??zsraw??o&namil?tainop,??z&eiwolaib?mol???c&e&iw&alselob?o&nsos?rtso???le&im?zrogz???orw,p??d&em,ia?ragrats?uolc&inu,sds,??e&c&i&lrog?w&ilg,o&hc&arats?orp??klop?tak????yzreibok??i&csjuoniws?ksromop?saldop??l&ahdop?opo??napokaz,t&atselaer?iselpmis,?z&romop?swozam???g&alble?ezrbo&lok?nrat??ro??hcyzrblaw?i&csomohcurein?grat?klawus??k&e&ru...

				{1, 23551}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R11 R12 R14
			},
			outputs: []outputInfo{
				{0, 23551}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R11 R12 R14
			},
		},
	},
	{
		name:         "ORW",
		argLen:       2,
		commutative:  true,
		clobberFlags: true,
		asm:          s390x.AORW,
		reg: regInfo{
			inputs: []inputInfo{
				{0, 23551}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R11 R12 R14