V4FMADDSS 7(AX), [X0-X3], K5, X30                  // 62627f0d9bb007000000 or 62627f2d9bb007000000 or 62627f4d9bb007000000
	V4FMADDSS (DI), [X0-X3], K5, X30                   // 62627f0d9b37 or 62627f2d9b37 or 62627f4d9b37
	V4FMADDSS 7(AX), [X10-X13], K5, X30                // 62622f0d9bb007000000 or 62622f2d9bb007000000 or 62622f4d9bb007000000
	V4FMADDSS (DI), [X10-X13], K5, X30                 // 62622f0d9b37 or 62622f2d9b37 or 62622f4d9b37

// func addMulVVW1024(z, x *uint, y uint) (c uint)
TEXT ·addMulVVW1024(SB),$0-32
	MOV	$16, X30
	JMP	addMulVVWx(SB)

// func addMulVVW1536(z, x *uint, y uint) (c uint)
TEXT ·addMulVVW1536(SB),$0-32
	MOV	$24, X30
	JMP	addMulVVWx(SB)

// func addMulVVW2048(z, x *uint, y uint) (c uint)
TEXT ·addMulVVW2048(SB),$0-32
	MOV	$32, X30
	JMP	addMulVVWx(SB)

TEXT addMulVVWx(SB),NOFRAME|NOSPLIT,$0
	MOV	z+0(FP), X5

// is computed with correct precision.
func truncate() {
	const (
		x30 = 1 << 30
		x60 = 1 << 60

		staticF32 = float32(x30) + 1 - x30
		staticF64 = float64(x60) + 1 - x60
		staticC64 = complex64(x30) + 1 - x30
		staticC128 = complex128(x60) + 1 - x60
	)
	dynamicF32 := float32(x30)
	dynamicF32 += 1
	dynamicF32 -= x30

	dynamicF64 := float64(x60)
	dynamicF64 += 1
	dynamicF64 -= x60

package main

func main() {
	var x00, x01, x02, x03, x04, x05, x06, x07, x08, x09 int
	var x10, x11, x12, x13, x14, x15, x16, x17, x18, x19 int
	var x20, x21, x22, x23, x24, x25, x26, x27, x28, x29 int
	var x30, x31, x32 int

	_ = x00
	_ = x01
	_ = x02
	_ = x03
	_ = x04
	_ = x05
	_ = x06
	_ = x07
	_ = x08
	_ = x09

	_ = x10
	_ = x11
	_ = x12
	_ = x13
	_ = x14
	_ = x15
	_ = x16

	//	_111111 = _111 + _111000
	//	x12     = _111111 << 6 + _111111
	//	x24     = x12 << 12 + x12
	//	x30     = x24 << 6 + _111111
	//	x31     = 2*x30 + 1
	//	x32     = 2*x31 + 1
	//	x63     = x32 << 31 + x31
	//	x126    = x63 << 63 + x63
	//	x252    = x126 << 126 + x126
	//	x255    = x252 << 3 + _111
	//	i397    = ((x255 << 33 + x32) << 94 + x30) << 2
	//	return    1 + i397
	//

	var z = new(P384Element).Set(e)

 * Calling into the gc tool chain, where all registers are caller save.
 * Called from standard ARM EABI, where x19-x29 are callee-save, so they
 * must be saved explicitly, along with x30 (LR).
 */
.globl EXT(crosscall1)
EXT(crosscall1):
	.cfi_startproc
	stp x29, x30, [sp, #-96]!
	.cfi_def_cfa_offset 96
	.cfi_offset 29, -96
	.cfi_offset 30, -88
	mov x29, sp
	.cfi_def_cfa_register 29
	stp x19, x20, [sp, #80]
	.cfi_offset 19, -16

	MOVBU	2(X12), X18
	MOVBU	3(X12), X20
	MOVBU	4(X12), X22
	MOVBU	5(X12), X24
	MOVBU	6(X12), X28
	MOVBU	7(X12), X30
	BNE	X8, X9, cmp1a
	BNE	X15, X16, cmp1b
	BNE	X17, X18, cmp1c
	BNE	X19, X20, cmp1d
	BNE	X21, X22, cmp1e
	BNE	X23, X24, cmp1f
	BNE	X25, X28, cmp1g
	BNE	X29, X30, cmp1h
	ADD	$8, X10
	ADD	$8, X12
	SUB	$8, X5
	BGE	X5, X6, compare8_unaligned
	BEQZ	X5, cmp_len

// This form typically matches the syntax defined in the ARM Reference Manual.
func GNUSyntax(inst Inst) string {
	switch inst.Op {
	case RET:
		if r, ok := inst.Args[0].(Reg); ok && r == X30 {
			return "ret"
		}
	case B:
		if _, ok := inst.Args[0].(Cond); ok {
			return strings.ToLower("b." + inst.Args[0].String() + " " + inst.Args[1].String())
		}
	case SYSL:
		result := strings.ToLower(inst.String())

	MOV X18, 112(X2)
	MOV X19, 120(X2)
	MOV X20, 128(X2)
	MOV X21, 136(X2)
	MOV X22, 144(X2)
	MOV X23, 152(X2)
	MOV X24, 160(X2)
	MOV X25, 168(X2)
	MOV X26, 176(X2)
	MOV X28, 184(X2)
	MOV X29, 192(X2)
	MOV X30, 200(X2)
	MOVD F0, 208(X2)
	MOVD F1, 216(X2)
	MOVD F2, 224(X2)
	MOVD F3, 232(X2)
	MOVD F4, 240(X2)
	MOVD F5, 248(X2)
	MOVD F6, 256(X2)
	MOVD F7, 264(X2)
	MOVD F8, 272(X2)
	MOVD F9, 280(X2)

	MSGSCHEDULE1(index); \
	SHA512ROUND(index, a, b, c, d, e, f, g, h)

// func block(dig *digest, p []byte)
TEXT ·block(SB),0,$128-32
	MOV	p_base+8(FP), X29
	MOV	p_len+16(FP), X30
	SRL	$7, X30
	SLL	$7, X30

	ADD	X29, X30, X28
	BEQ	X28, X29, end

	MOV	·_K(SB), X18		// const table
	ADD	$8, X2, X19		// message schedule

	MOV	dig+0(FP), X20
	MOV	(0*8)(X20), X10		// a = H0