VPCONFLICTD X6, K6, X6                             // 62f27d0ec4f6
	VPCONFLICTD X1, K6, X6                             // 62f27d0ec4f1
	VPCONFLICTD X8, K6, X6                             // 62d27d0ec4f0
	VPCONFLICTD 15(R8), K6, X6                         // 62d27d0ec4b00f000000
	VPCONFLICTD (BP), K6, X6                           // 62f27d0ec47500
	VPCONFLICTD X6, K6, X17                            // 62e27d0ec4ce

// func archMax(x, y float64) float64
TEXT ·archMax(SB),NOSPLIT,$0
	MOVD	x+0(FP), F0
	MOVD	y+8(FP), F1
	FCLASSD	F0, X5
	FCLASSD	F1, X6

	// +Inf special cases
	MOV	$PosInf, X7
	BEQ	X7, X5, isMaxX
	BEQ	X7, X6, isMaxY

	// NaN special cases
	MOV	$NaN, X7
	BEQ	X7, X5, isMaxX
	BEQ	X7, X6, isMaxY

	// normal case
	FMAXD	F0, F1, F0
	MOVD	F0, ret+16(FP)
	RET

isMaxX: // return x
	MOVD	F0, ret+16(FP)
	RET

		x4 := s[1]
		x5 := s[7]
		x6 := s[5]
		x7 := s[3]

		// Stage 1.
		x8 := w7 * (x4 + x5)
		x4 = x8 + w1mw7*x4
		x5 = x8 - w1pw7*x5
		x8 = w3 * (x6 + x7)
		x6 = x8 - w3mw5*x6
		x7 = x8 - w3pw5*x7

		// Stage 2.
		x8 = x0 + x1
		x0 -= x1
		x1 = w6 * (x3 + x2)
		x2 = x1 - w2pw6*x2
		x3 = x1 + w2mw6*x3
		x1 = x4 + x6
		x4 -= x6
		x6 = x5 + x7
		x5 -= x7

	ADD	$(1024*8), X2

	// Ask signaller to setgid
	MOV	$1, X5
	FENCE
	MOVW	X5, ·Baton(SB)
	FENCE

	// Wait for setgid completion
loop:
	FENCE
	MOVW	·Baton(SB), X5
	OR	X6, X6, X6	// hint that we're in a spin loop
	BNE	ZERO, X5, loop
	FENCE

	// Restore stack
	ADD	$(-1024*8), X2

const x5 = 1p10   // ERROR "'p' exponent requires hexadecimal mantissa|invalid prefix"
const x6 = 0P0    // ERROR "'P' exponent requires hexadecimal mantissa|invalid prefix"

func main() {
	fmt.Printf("%g %T\n", x1, x1)
	fmt.Printf("%g %T\n", x2, x2)
	fmt.Printf("%g %T\n", x3, x3)
	fmt.Printf("%g %T\n", x4, x4)
	fmt.Printf("%g %T\n", x5, x5)
	fmt.Printf("%g %T\n", x6, x6)

	VGATHERDPS X3, 664(X4*8), X6            // c4e2619234e598020000
	VGATHERDPS X3, 8(X14*1), X6             // c4a26192343508000000
	VGATHERDPS X3, -8(X14*1), X6            // c4a261923435f8ffffff
	VGATHERDPS X3, 0(X14*1), X6             // c4a26192343500000000
	VGATHERDPS X3, 664(X14*1), X6           // c4a26192343598020000
	VGATHERDPS X3, 8(X14*8), X6             // c4a2619234f508000000
	VGATHERDPS X3, -8(X14*8), X6            // c4a2619234f5f8ffffff

	VPDPBUSD X1, X28, K2, X6                           // 62f21d0250f1
	VPDPBUSD -15(R14)(R15*1), X28, K2, X6              // 62921d0250b43ef1ffffff
	VPDPBUSD -15(BX), X28, K2, X6                      // 62f21d0250b3f1ffffff
	VPDPBUSD X15, X8, K2, X6                           // 62d23d0a50f7
	VPDPBUSD X11, X8, K2, X6                           // 62d23d0a50f3

	VREPF $0, KEY0, X4
	VREPF $1, KEY0, X5
	VREPF $2, KEY0, X6
	VREPF $3, KEY0, X7
	VREPF $0, KEY1, X8
	VREPF $1, KEY1, X9
	VREPF $2, KEY1, X10
	VREPF $3, KEY1, X11
	VLR   CTR, X12
	VREPF $1, NONCE, X13
	VREPF $2, NONCE, X14
	VREPF $3, NONCE, X15

	MOVD $(NUM_ROUNDS/2), R1

loop:
	ROUND4(X0, X4, X12,  X8, X1, X5, X13,  X9, X2, X6, X14, X10, X3, X7, X15, X11)

	JALR	RA, (X5)

	// Restore g, stack pointer. X10 is return value.
	MOV	0(X2), g
	CALL	runtime·save_g(SB)
	MOV	(g_stack+stack_hi)(g), X5
	MOV	8(X2), X6
	SUB	X6, X5, X6
	MOV	X6, X2

	MOVW	X10, ret+16(FP)
	RET

// func asminit()
TEXT runtime·asminit(SB),NOSPLIT|NOFRAME,$0-0
	RET

// reflectcall: call a function with the given argument list

	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)
	RET
move_129through256:
	MOVOU	(SI), X0
	MOVOU	16(SI), X1
	MOVOU	32(SI), X2