MOVQ	s2+24(FP), RARG2
	MOVQ	result+32(FP), RARG3

	get_tls(R12)
	MOVQ	g(R12), R14
	MOVQ	g_m(R14), R13

	// Switch to g0 stack.
	MOVQ	SP, R12		// callee-saved, preserved across the CALL
	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
	RET

	"VPCMPEQQ",
	"VPCMPEQW",
	"VPCMPESTRI",
	"VPCMPESTRM",
	"VPCMPGTB",
	"VPCMPGTD",
	"VPCMPGTQ",
	"VPCMPGTW",
	"VPCMPISTRI",
	"VPCMPISTRM",
	"VPCMPQ",
	"VPCMPUB",
	"VPCMPUD",
	"VPCMPUQ",
	"VPCMPUW",
	"VPCMPW",
	"VPCOMPRESSB",
	"VPCOMPRESSD",
	"VPCOMPRESSQ",
	"VPCOMPRESSW",
	"VPCONFLICTD",
	"VPCONFLICTQ",
	"VPDPBUSD",
	"VPDPBUSDS",
	"VPDPWSSD",

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:

MOVQ $name+10(SB), AX -> MOVQ name@GOT(SB), AX; LEAQ 10(AX), AX

MOVQ name(SB), AX -> NOP; MOVQ name@GOT(SB), R15; MOVQ (R15), AX
MOVQ name+10(SB), AX -> NOP; MOVQ name@GOT(SB), R15; MOVQ 10(R15), AX

CMPQ name(SB), $0 -> NOP; MOVQ name@GOT(SB), R15; CMPQ (R15), $0

MOVQ $1, name(SB) -> NOP; MOVQ name@GOT(SB), R15; MOVQ $1, (R15)
MOVQ $1, name+10(SB) -> NOP; MOVQ name@GOT(SB), R15; MOVQ $1, 10(R15)
`

type ParsedTestData struct {

			// 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

		case STOSB, MOVSB, CMPSB, LODSB, SCASB:
			prefix = "byte "
		case STOSW, MOVSW, CMPSW, LODSW, SCASW:
			prefix = "word "
		case STOSD, MOVSD, CMPSD, LODSD, SCASD:
			prefix = "dword "
		case STOSQ, MOVSQ, CMPSQ, LODSQ, SCASQ:
			prefix = "qword "
		case LAR:
			prefix = "word "
		case BOUND:
			if inst.Mode == 32 {
				prefix = "qword "
			} else {
				prefix = "dword "
			}

	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