FXCHD   F0, F1       // F0=q (smaller), F1=p (larger)
	FDIVD   F1, F0       // F0=q(=q/p), F1=p
	FMULD   F0, F0       // F0=q*q, F1=p
	FLD1                 // F0=1, F1=q*q, F2=p
	FADDDP  F0, F1       // F0=1+q*q, F1=p
	FSQRT                // F0=sqrt(1+q*q), F1=p
	FMULDP  F0, F1       // F0=p*sqrt(1+q*q)
	FMOVDP  F0, ret+16(FP)
	RET
	FMOVDP  F0, F1       // F0=0
	FMOVDP  F0, ret+16(FP)
	RET
not_finite:
// test bits for -Inf or +Inf

	FMOVD	200(R9), F2
	FMADD	F2, F0, F0
L1:
	FMOVD	F0, ret+8(FP)
	RET
L14:
	LTDBR	F0, F0
	BLTU	L17
	FMOVD	F0, F10
L4:
	FMOVD	192(R9), F2
	WFMADB	V0, V0, V2, V0
	LTDBR	F0, F0
	FSQRT	F0, F8
L5:
	WFADB	V8, V10, V0
	WORD	$0xC0398006	//iilf	%r3,2147909631
	BYTE	$0x7F
	BYTE	$0xFF
	LGDR	F0, R5
	SRAD	$32, R5
	MOVH	$0x0, R2
	SUBW	R5, R3
	RISBGZ	$32, $47, $0, R3, R4

	FMOVD	0(R3)(R1*1), F3
	WFMADB	V0, V4, V3, V0
	FMOVD	8(R9), F4
	FMADD	F4, F2, F0
	FMOVD	F0, ret+8(FP)
	RET
L10:
	FMOVD	F0, F8
	FMOVD	0(R9), F0
	FMADD	F8, F8, F0
	LTDBR	F0, F0
	FSQRT	F0, F10
L4:
	WFADB	V10, V8, V0
	WORD	$0xC0398006	//iilf	%r3,2147909631
	BYTE	$0x7F
	BYTE	$0xFF
	LGDR	F0, R5
	SRAD	$32, R5
	MOVH	$0x0, R1
	SUBW	R5, R3
	SRAW	$8, R3, R2

	"FCPSGNCC",
	"FRES",
	"FRESCC",
	"FRIM",
	"FRIMCC",
	"FRIP",
	"FRIPCC",
	"FRIZ",
	"FRIZCC",
	"FRIN",
	"FRINCC",
	"FRSQRTE",
	"FRSQRTECC",
	"FSEL",
	"FSELCC",
	"FSQRT",
	"FSQRTCC",
	"FSQRTS",
	"FSQRTSCC",
	"CNTLZD",
	"CNTLZDCC",
	"CMPW",
	"CMPWU",
	"CMPB",
	"FTDIV",
	"FTSQRT",
	"DIVD",
	"DIVDCC",
	"DIVDE",
	"DIVDECC",
	"DIVDEU",

	BVS	L1
	FMOVD	16(R9), F10
	BR	L1
L12:
	FMOVD	24(R9), F0
	FMADD	F10, F10, F0
	WORD	$0xB3130080	//lcdbr	%f8,%f0
	WORD	$0xED009008	//cdb	%f0,.L37-.L13(%r9)
	BYTE	$0x00
	BYTE	$0x19
	FSQRT	F8, F10
L5:
	MOVW	R12, R4
	CMPBLE	R4, $0, L7
	WORD	$0xB31300AA	//lcdbr	%f10,%f10
	FMOVD	$0, F1
	BR	L3
L9:
	FMOVD	0(R9), F10
	BR	L1
L8:
	FMOVD	$0, F0
	FMOVD	F0, ret+8(FP)
	RET

	FMADD	F0, F1, F8
	FMOVD	F8, ret+8(FP)
	RET
L14:
	FMOVD	200(R9), F0
	FMADD	F8, F8, F0
	WORD	$0xB31300A0	//lcdbr	%f10,%f0
	WORD	$0xED009020	//cdb	%f0,.L39-.L15(%r9)
	BYTE	$0x00
	BYTE	$0x19
	FSQRT	F10, F8
L6:
	MOVW	R7, R6
	CMPBLE	R6, $0, L8
	WORD	$0xB3130088	//lcdbr	%f8,%f8
	FMOVD	24(R9), F1
	BR	L4
L10:
	FMOVD	16(R9), F8
	BR	L1
L9:
	FMOVD	8(R9), F8
	FMOVD	F8, ret+8(FP)
	RET

	"FMADD",
	"FMADDS",
	"FMOVD",
	"FMOVS",
	"FMSUB",
	"FMSUBS",
	"FMUL",
	"FMULS",
	"FNABS",
	"FNEG",
	"FNEGS",
	"LEDBR",
	"LDEBR",
	"LPDFR",
	"LNDFR",
	"FSUB",
	"FSUBS",
	"FSQRT",
	"FSQRTS",
	"FIEBR",
	"FIDBR",
	"CPSDR",
	"LTEBR",
	"LTDBR",
	"TCEB",
	"TCDB",
	"LDGR",
	"LGDR",
	"CEFBRA",
	"CDFBRA",
	"CEGBRA",
	"CDGBRA",
	"CFEBRA",
	"CFDBRA",

	// 386/sse2:"SQRTSD" 386/softfloat:-"SQRTD"
	// arm64:"FSQRTD"
	// arm/7:"SQRTD"
	// mips/hardfloat:"SQRTD" mips/softfloat:-"SQRTD"
	// mips64/hardfloat:"SQRTD" mips64/softfloat:-"SQRTD"
	// wasm:"F64Sqrt"
	// ppc64x:"FSQRT"
	// riscv64: "FSQRTD"
	return math.Sqrt(x)
}

func sqrt32(x float32) float32 {
	// amd64:"SQRTSS"
	// 386/sse2:"SQRTSS" 386/softfloat:-"SQRTS"
	// arm64:"FSQRTS"
	// arm/7:"SQRTF"

			args = nil
			op = "int3"
		}

	case LCALL, LJMP:
		if len(args) == 2 {
			args[0], args[1] = args[1], args[0]
		}

	case FCHS, FABS, FTST, FLDPI, FLDL2E, FLDLG2, F2XM1, FXAM, FLD1, FLDL2T, FSQRT, FRNDINT, FCOS, FSIN:
		if len(args) == 0 {
			args = append(args, "st0")
		}

	case FPTAN, FSINCOS, FUCOMPP, FCOMPP, FYL2X, FPATAN, FXTRACT, FPREM1, FPREM, FYL2XP1, FSCALE:
		if len(args) == 0 {

	"FMULDP",
	"FMULF",
	"FMULL",
	"FMULW",
	"FNOP",
	"FPATAN",
	"FPREM",
	"FPREM1",
	"FPTAN",
	"FRNDINT",
	"FRSTOR",
	"FSAVE",
	"FSCALE",
	"FSIN",
	"FSINCOS",
	"FSQRT",
	"FSTCW",
	"FSTENV",
	"FSTSW",
	"FSUBD",
	"FSUBDP",
	"FSUBF",
	"FSUBL",
	"FSUBRD",
	"FSUBRDP",
	"FSUBRF",
	"FSUBRL",
	"FSUBRW",
	"FSUBW",
	"FTST",
	"FUCOM",