"VCLZB",
	"VCLZH",
	"VCLZW",
	"VCLZD",
	"VPOPCNT",
	"VPOPCNTB",
	"VPOPCNTH",
	"VPOPCNTW",
	"VPOPCNTD",
	"VCMPEQ",
	"VCMPEQUB",
	"VCMPEQUBCC",
	"VCMPEQUH",
	"VCMPEQUHCC",
	"VCMPEQUW",
	"VCMPEQUWCC",
	"VCMPEQUD",
	"VCMPEQUDCC",
	"VCMPGT",
	"VCMPGTUB",
	"VCMPGTUBCC",
	"VCMPGTUH",
	"VCMPGTUHCC",
	"VCMPGTUW",
	"VCMPGTUWCC",
	"VCMPGTUD",
	"VCMPGTUDCC",

TEXT ·xorBytes(SB), NOSPLIT, $0
	MOVD	dst+0(FP), R3	// R3 = dst
	MOVD	a+8(FP), R4	// R4 = a
	MOVD	b+16(FP), R5	// R5 = b
	MOVD	n+24(FP), R6	// R6 = n

	CMPU	R6, $64, CR7	// Check if n ≥ 64 bytes
	MOVD	R0, R8		// R8 = index
	CMPU	R6, $8, CR6	// Check if 8 ≤ n < 64 bytes
	BLE	CR6, small	// <= 8
	BLT	CR7, xor32	// Case for 32 ≤ n < 64 bytes

	// Case for n ≥ 64 bytes
preloop64:

			opset(AVPOPCNTD, r0)

		case AVCMPEQ: /* vcmpequb[.], vcmpequh[.], vcmpequw[.], vcmpequd[.] */
			opset(AVCMPEQUB, r0)
			opset(AVCMPEQUBCC, r0)
			opset(AVCMPEQUH, r0)
			opset(AVCMPEQUHCC, r0)
			opset(AVCMPEQUW, r0)
			opset(AVCMPEQUWCC, r0)
			opset(AVCMPEQUD, r0)
			opset(AVCMPEQUDCC, r0)

		case AVCMPGT: /* vcmpgt[u,s]b[.], vcmpgt[u,s]h[.], vcmpgt[u,s]w[.], vcmpgt[u,s]d[.] */

			// guard against underflow.
			//
			//	CMPU	SP, $(framesize-StackSmall)
			//	BLT	label-of-call-to-morestack
			if offset <= 0xffff {
				p = obj.Appendp(p, c.newprog)
				p.As = ACMPU
				p.From.Type = obj.TYPE_REG
				p.From.Reg = REGSP
				p.To.Type = obj.TYPE_CONST
				p.To.Offset = offset
			} else {
				// Constant is too big for CMPU.
				p = obj.Appendp(p, c.newprog)
				p.As = AMOVD

	return z - x*y
}

func Cmp(f float64) bool {
	// arm64:"FCMPD","(BGT|BLE|BMI|BPL)",-"CSET\tGT",-"CBZ"
	return f > 4 || f < -4
}

func CmpZero64(f float64) bool {
	// s390x:"LTDBR",-"FCMPU"
	return f <= 0
}

func CmpZero32(f float32) bool {
	// s390x:"LTEBR",-"CEBR"
	return f <= 0
}

func CmpWithSub(a float64, b float64) bool {
	f := a - b
	// s390x:-"LTDBR"
	return f <= 0

	AVSRD
	AVSA
	AVSRAB
	AVSRAH
	AVSRAW
	AVSRAD
	AVSOI
	AVSLDOI
	AVCLZ
	AVCLZB
	AVCLZH
	AVCLZW
	AVCLZD
	AVPOPCNT
	AVPOPCNTB
	AVPOPCNTH
	AVPOPCNTW
	AVPOPCNTD
	AVCMPEQ
	AVCMPEQUB
	AVCMPEQUBCC
	AVCMPEQUH
	AVCMPEQUHCC
	AVCMPEQUW
	AVCMPEQUWCC
	AVCMPEQUD
	AVCMPEQUDCC
	AVCMPGT
	AVCMPGTUB
	AVCMPGTUBCC
	AVCMPGTUH
	AVCMPGTUHCC
	AVCMPGTUW

	CMP R3, R4                      // 7c232000
	CMP R3, R0                      // 7c230000
	CMP R3, R0, CR1                 // CMP R3,CR1,R0   // 7ca30000
	CMPU R3, R4                     // 7c232040
	CMPU R3, R0                     // 7c230040
	CMPU R3, R0, CR2                // CMPU R3,CR2,R0  // 7d230040
	CMPW R3, R4                     // 7c032000
	CMPW R3, R0                     // 7c030000

	// The goal here is stability in the face
	// of unrelated changes elsewhere in the compiler.
	if c := x.AuxInt - y.AuxInt; c != 0 {
		return c < 0
	}
	if cmp := x.Type.Compare(y.Type); cmp != types.CMPeq {
		return cmp == types.CMPlt
	}
	return x.ID < y.ID
}

func (op Op) isLoweredGetClosurePtr() bool {
	switch op {

(Load <t1> p1 (Store {t2} p2 x _))
	&& isSamePtr(p1, p2)
	&& t1.Compare(x.Type) == types.CMPeq
	&& t1.Size() == t2.Size()
	=> x
(Load <t1> p1 (Store {t2} p2 _ (Store {t3} p3 x _)))
	&& isSamePtr(p1, p3)
	&& t1.Compare(x.Type) == types.CMPeq
	&& t1.Size() == t2.Size()
	&& disjoint(p3, t3.Size(), p2, t2.Size())
	=> x

	// population count
	{i: 9, as: APOPCNT, a1: C_REG, a6: C_REG},

	// compare
	{i: 70, as: ACMP, a1: C_REG, a6: C_REG},
	{i: 71, as: ACMP, a1: C_REG, a6: C_LCON},
	{i: 70, as: ACMPU, a1: C_REG, a6: C_REG},
	{i: 71, as: ACMPU, a1: C_REG, a6: C_LCON},
	{i: 70, as: AFCMPO, a1: C_FREG, a6: C_FREG},
	{i: 70, as: AFCMPO, a1: C_FREG, a2: C_REG, a6: C_FREG},

	// test under mask
	{i: 91, as: ATMHH, a1: C_REG, a6: C_ANDCON},