"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",

	PCALIGN $16
bigloop:
	CMPQ	BX, $8
	JBE	leftover
	MOVQ	(SI), CX
	MOVQ	(DI), DX
	ADDQ	$8, SI
	ADDQ	$8, DI
	SUBQ	$8, BX
	CMPQ	CX, DX
	JEQ	bigloop
	XORQ	AX, AX	// return 0
	RET

	// remaining 0-8 bytes
leftover:
	MOVQ	-8(SI)(BX*1), CX
	MOVQ	-8(DI)(BX*1), DX
	CMPQ	CX, DX
	SETEQ	AX
	RET

small:
	CMPQ	BX, $0
	JEQ	equal

	LEAQ	0(BX*8), CX

	VEOR	V7.B8, V7.B8, V7.B8
	VEOR	V8.B8, V8.B8, V8.B8
	PCALIGN $16
	// Count the target byte in 32-byte chunk
chunk_loop:
	VLD1.P	(R0), [V1.B16, V2.B16]
	CMP	R0, R3
	VCMEQ	V0.B16, V1.B16, V3.B16
	VCMEQ	V0.B16, V2.B16, V4.B16
	// Clear the higher 7 bits
	VAND	V5.B16, V3.B16, V3.B16
	VAND	V5.B16, V4.B16, V4.B16
	// Count lanes match the requested byte
	VADDP	V4.B16, V3.B16, V6.B16 // 32B->16B

TEXT	msancall<>(SB), NOSPLIT, $0-0
	get_tls(R12)
	MOVQ	g(R12), R14
	MOVQ	SP, R12		// callee-saved, preserved across the CALL
	CMPQ	R14, $0
	JE	call	// no g; still on a system stack

	MOVQ	g_m(R14), R13
	// Switch to g0 stack.
	MOVQ	m_g0(R13), R10
	CMPQ	R10, R14
	JE	call	// already on g0

	MOVQ	(g_sched+gobuf_sp)(R10), SP
call:
	ANDQ	$~15, SP	// alignment for gcc ABI
	CALL	AX
	MOVQ	R12, SP

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:

			// the spill location.
			var name LocalSlot
			if v.Op == OpStoreReg {
				name = names[v.Args[0].ID]
			} else {
				name = names[v.ID]
			}
			if name.N != nil && v.Type.Compare(name.Type) == types.CMPeq {
				for _, id := range s.interfere[v.ID] {
					h := f.getHome(id)
					if h != nil && h.(LocalSlot).N == name.N && h.(LocalSlot).Off == name.Off {
						// A variable can interfere with itself.

			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

	MOVL R15, (7*4)(BP)

	ADDQ $64, SI
	CMPQ SI, 256(SP)
	JB   loop

end:
	RET

avx2:
	MOVQ dig+0(FP), CTX          // d.h[8]
	MOVQ p_base+8(FP), INP
	MOVQ p_len+16(FP), NUM_BYTES

	LEAQ -64(INP)(NUM_BYTES*1), NUM_BYTES // Pointer to the last block
	MOVQ NUM_BYTES, _INP_END(SP)

	CMPQ NUM_BYTES, INP
	JE   avx2_only_one_block

	// Load initial digest