SHUFPS $0x10, X0, X4
	PXOR X4, X0
	SHUFPS $0x8c, X0, X4
	PXOR X4, X0
	PXOR X1, X0

	MOVAPS X2, X5
	MOVAPS X2, X6
	PSLLDQ $0x4, X5
	PSHUFD $0xff, X0, X3
	PXOR X3, X2
	PXOR X5, X2

	MOVAPS X0, X1
	SHUFPS $0x44, X0, X6
	MOVUPS X6, (BX)
	SHUFPS $0x4e, X2, X1
	MOVUPS X1, 16(BX)
	ADDQ $32, BX
	RET

TEXT _expand_key_192b<>(SB),NOSPLIT,$0
	PSHUFD $0x55, X1, X1

	MOVOU	16(SI), X1
	MOVOU	32(SI), X2
	MOVOU	48(SI), X3
	MOVOU	-64(SI)(BX*1), X4
	MOVOU	-48(SI)(BX*1), X5
	MOVOU	-32(SI)(BX*1), X6
	MOVOU	-16(SI)(BX*1), X7
	MOVOU	X0, (DI)
	MOVOU	X1, 16(DI)
	MOVOU	X2, 32(DI)
	MOVOU	X3, 48(DI)
	MOVOU	X4, -64(DI)(BX*1)
	MOVOU	X5, -48(DI)(BX*1)
	MOVOU	X6, -32(DI)(BX*1)
	MOVOU	X7, -16(DI)(BX*1)

var x5 /* ERROR "initialization cycle" */ = T5{}.m() // <<<< added ()

var y5 = x5

type T5 struct{}

func (T5) m() int {
	_ = y5
	return 0
}

// issue 4847
// simplified test case

var x6 = f6
var y6 /* ERROR "initialization cycle" */ = f6
func f6() { _ = y6 }

// full test case

type (
      E int
      S int
)

type matcher func(s *S) E

var vx3 /* ERROR "initialization cycle: vx3 refers to itself" */ S1 = S1{vx3.f}

// cycles via functions

var x4 = x5
var x5 /* ERROR "initialization cycle for x5" */ = f1()
func f1() int { return x5*10 }

var x6, x7 /* ERROR "initialization cycle" */ = f2()
var x8 = x7
func f2() (int, int) { return f3() + f3(), 0 }
func f3() int { return x8 }

// cycles via function literals

	TEQ	R1, R5
	BNE Y7

X7:
	MOVW	$0, R1
	MOVW	R1, c+28(FP)
	RET


// func shrVU(z, x []Word, s uint) (c Word)
TEXT ·shrVU(SB),NOSPLIT,$0
	MOVW	z_len+4(FP), R5
	TEQ	$0, R5
	BEQ	X6

	MOVW	z+0(FP), R1
	MOVW	x+12(FP), R2
	ADD	R5<<2, R1, R5
	MOVW	s+24(FP), R3
	TEQ	$0, R3	// shift 0 is special
	BEQ Y6
	SUB	$4, R5	// stop one word early
	MOVW	$32, R4
	SUB	R3, R4
	MOVW	$0, R7

	M1:   "M1",
	M2:   "M2",
	M3:   "M3",
	M4:   "M4",
	M5:   "M5",
	M6:   "M6",
	M7:   "M7",
	X0:   "X0",
	X1:   "X1",
	X2:   "X2",
	X3:   "X3",
	X4:   "X4",
	X5:   "X5",
	X6:   "X6",
	X7:   "X7",
	X8:   "X8",
	X9:   "X9",
	X10:  "X10",
	X11:  "X11",
	X12:  "X12",
	X13:  "X13",
	X14:  "X14",
	X15:  "X15",
	CS:   "CS",
	SS:   "SS",
	DS:   "DS",
	ES:   "ES",

	p256OrdSqr(t, x, 1)          // _10100
	p256OrdMul(_10101, t, _1)    // _10101
	p256OrdSqr(x, _10101, 1)     // _101010
	p256OrdMul(_101111, _101, x) // _101111
	p256OrdMul(x, _10101, x)     // _111111 = x6
	p256OrdSqr(t, x, 2)          // _11111100
	p256OrdMul(t, t, _11)        // _11111111 = x8
	p256OrdSqr(x, t, 8)          // _ff00
	p256OrdMul(x, x, t)          // _ffff = x16
	p256OrdSqr(t, x, 16)         // _ffff0000

	// seq 1 206 | sed 's/.*/	var x& [10<<20]byte/'
	// seq 1 206 | sed 's/.*/	z = x&/'
	var x1 [10<<20]byte
	var x2 [10<<20]byte
	var x3 [10<<20]byte
	var x4 [10<<20]byte
	var x5 [10<<20]byte
	var x6 [10<<20]byte
	var x7 [10<<20]byte
	var x8 [10<<20]byte
	var x9 [10<<20]byte
	var x10 [10<<20]byte
	var x11 [10<<20]byte
	var x12 [10<<20]byte
	var x13 [10<<20]byte
	var x14 [10<<20]byte
	var x15 [10<<20]byte

instructions and floating-point(scalar) instructions.

Examples:

	VADD V5.H8, V18.H8, V9.H8         <=>      add v9.8h, v18.8h, v5.8h
	VLD1.P (R6)(R11), [V31.D1]        <=>      ld1 {v31.1d}, [x6], x11
	VFMLA V29.S2, V20.S2, V14.S2      <=>      fmla v14.2s, v20.2s, v29.2s
	AESD V22.B16, V19.B16             <=>      aesd v19.16b, v22.16b
	SCVTFWS R3, F16                   <=>      scvtf s17, w6

6. Align directive