// 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

	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

	CTZ	X24, X25				// 931c1c60
	CTZW	X25, X26				// 1b9d1c60
	MAX	X26, X28, X29				// b36eae0b
	MAX	X26, X28				// 336eae0b
	MAXU	X28, X29, X30				// 33ffce0b
	MAXU	X28, X29				// b3fece0b
	MIN	X29, X30, X5				// b342df0b
	MIN	X29, X30				// 334fdf0b
	MINU	X30, X5, X6				// 33d3e20b
	MINU	X30, X5					// b3d2e20b
	ORN	X6, X7, X8				// 33e46340
	ORN	X6, X7					// b3e36340
	SEXTB	X16, X17				// 93184860

	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

				{2, 1073741808},          // X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 g X28 X29 X30
				{0, 9223372037928517618}, // SP X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 g X28 X29 X30 SB
			},
			outputs: []outputInfo{
				{0, 1006632944}, // X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X28 X29 X30
			},
		},
	},

	var x21 [10<<20]byte
	var x22 [10<<20]byte
	var x23 [10<<20]byte
	var x24 [10<<20]byte
	var x25 [10<<20]byte
	var x26 [10<<20]byte
	var x27 [10<<20]byte
	var x28 [10<<20]byte
	var x29 [10<<20]byte
	var x30 [10<<20]byte
	var x31 [10<<20]byte
	var x32 [10<<20]byte
	var x33 [10<<20]byte
	var x34 [10<<20]byte
	var x35 [10<<20]byte
	var x36 [10<<20]byte
	var x37 [10<<20]byte
	var x38 [10<<20]byte
	var x39 [10<<20]byte

	exc uint32 // number of arm exception taken
}

type regs64 struct {
	x     [29]uint64 // registers x0 to x28
	fp    uint64     // frame register, x29
	lr    uint64     // link register, x30
	sp    uint64     // stack pointer, x31
	pc    uint64     // program counter
	cpsr  uint32     // current program status register
	__pad uint32
}

type neonstate64 struct {

    X13,
    X14,
    X15,
    X16,
    X17,
    X18,
    X19,
    X20,
    X21,
    X22,
    X23,
    X24,
    X25,
    X26,
    X27,
    X28,
    X29,
    X30,
    X31,
    X32,
    X33,
    X34,
    X35,
    X36,
    X37,
    X38,
    X39,
    X40,
    X41,
    X42,
    X43,
    X44,
    X45,
    X46,
    X47,

	X5
	X6
	X7
	X8
	X9
	X10
	X11
	X12
	X13
	X14
	X15
	X16
	X17
	X18
	X19
	X20
	X21
	X22
	X23
	X24
	X25
	X26
	X27
	X28
	X29
	X30
	XZR

	B0
	B1
	B2
	B3
	B4
	B5
	B6
	B7
	B8
	B9
	B10
	B11
	B12
	B13
	B14
	B15
	B16
	B17
	B18
	B19
	B20
	B21
	B22
	B23