VSEL     V14, V13, V31, V7  // final merge
	VCLZD    V7, V18            // Find first index for each half
	MFVSRD   V18, R25           // Isolate value
	CMP      R25, $64           // If < 64, found
	BLT      foundR25           // Return found index
	VSLDOI   $8, V18, V18, V18  // Move for MFVSRD
	MFVSRD   V18, R25           // Isolate other value
	CMP      R25, $64           // If < 64, found

	for i := 0; i <= 31; i++ {
		reg := fmt.Sprintf("F%d", i)
		l.add("FMOVD", reg, 8)
	}
	// Add floating point control/status register FPSCR.
	l.addSpecial(
		"MOVFL FPSCR, F0\nFMOVD F0, %d(R1)",
		"FMOVD %d(R1), F0\nMOVFL F0, FPSCR",
		8)

	p("MOVD R31, -%d(R1)", l.stack-32) // save R31 first, we'll use R31 for saving LR
	p("MOVD LR, R31")

//		outcode(int($1), &$2, 0, &$4);
//	}
	MOVD	F1, foo<>+3(SB)
	MOVD	F1, 16(R2)
	MOVD	F1, (R2)

//
// floating point status
//
//	LMOVW fpscr ',' freg
//	{
//		outcode(int($1), &$2, 0, &$4);
//	}
	MOVW	FCR31, R1 // 4441f800

//	LMOVW freg ','  fpscr
//	{
//		outcode(int($1), &$2, 0, &$4);
//	}
	MOVW	R1, FCR31 // 44c1f800

//	LMOVW rreg ',' mreg
//	{
//		outcode(int($1), &$2, 0, &$4);

	LFSU: "FMOVSU", STFSU: "FMOVSU",
	LFSUX: "FMOVSU", STFSUX: "FMOVSU",
	CMPD: "CMP", CMPDI: "CMP",
	CMPW: "CMPW", CMPWI: "CMPW",
	CMPLD: "CMPU", CMPLDI: "CMPU",
	CMPLW: "CMPWU", CMPLWI: "CMPWU",
	MTSPR: "MOVD", MFSPR: "MOVD", // the width is ambiguous for SPRs
	B:  "BR",
	BL: "CALL",

// It returns nil if x cannot be decoded according to aop.
func decodeArg(aop instArg, x uint32) Arg {
	switch aop {
	default:
		return nil

	case arg_APSR:
		return APSR
	case arg_FPSCR:
		return FPSCR

	case arg_R_0:
		return Reg(x & (1<<4 - 1))
	case arg_R_8:
		return Reg((x >> 8) & (1<<4 - 1))
	case arg_R_12:
		return Reg((x >> 12) & (1<<4 - 1))
	case arg_R_16:
		return Reg((x >> 16) & (1<<4 - 1))

				startArg = 1
			case 9:
				buf.WriteString("ctr")
				startArg = 1
			default:
				buf.WriteString("spr")
			}
		default:
			buf.WriteString("spr")
		}

	case "mfspr":
		opcode := inst.Op.String()
		buf.WriteString(opcode[0:2])
		arg := inst.Args[0]
		switch spr := inst.Args[1].(type) {
		case SpReg:
			switch spr {
			case 1:
				buf.WriteString("xer ")

	C_SOREG    /* An $n+reg memory arg where n is a 16 bit signed offset */
	C_LOREG    /* An $n+reg memory arg where n is a 32 bit signed offset */
	C_XOREG    /* An reg+reg memory arg */
	C_FPSCR    /* The fpscr register */
	C_LR       /* The link register */
	C_CTR      /* The count register */
	C_ANY      /* Any argument */
	C_GOK      /* A non-matched argument */
	C_ADDR     /* A symbolic memory location */

	VSLDOI	$8,V0,zeroes,V0

#else

	VAND	V0,mask_32bit,V1
	VPMSUMD	V1,const1,V1
	VAND	V1,mask_32bit,V1
	VPMSUMD	V1,const2,V1
	VXOR	V0,V1,V0
	VSLDOI  $4,V0,zeroes,V0

#endif

	MFVSRD	VS32,R3 // VS32 = V0

	NOR	R3,R3,R3 // return ^crc
	MOVW	R3,ret+32(FP)
	RET

first_warm_up_done:

	LVX	(R3),const1
	ADD	$16,R3

	VPMSUMD	V16,const1,V8
	VPMSUMD	V17,const1,V9

		// There are optimizations that should apply -- (Xi2f64 (MOVWload (not-ADD-ptr+offset) ) ) could use
		// the word-load instructions.  (Xi2f64 (MOVDload ptr )) can be (FMOVDload ptr)

		{name: "MFVSRD", argLength: 1, reg: fpgp, asm: "MFVSRD", typ: "Int64"},   // move 64 bits of F register into G register
		{name: "MTVSRD", argLength: 1, reg: gpfp, asm: "MTVSRD", typ: "Float64"}, // move 64 bits of G register into F register

	// disable runfast (flush-to-zero) mode of vfp if runtime.goarmsoftfp == 0
	MOVB	runtime·goarmsoftfp(SB), R11
	CMP	$0, R11
	BNE	4(PC)
	WORD	$0xeef1ba10	// vmrs r11, fpscr
	BIC	$(1<<24), R11
	WORD	$0xeee1ba10	// vmsr fpscr, r11
	RET

TEXT runtime·mstart(SB),NOSPLIT|TOPFRAME,$0
	BL	runtime·mstart0(SB)
	RET // not reached

/*
 *  go-routine
 */

// void gogo(Gobuf*)