VCMEQ V24.S4, V13.S4, V12.S4                                // ac8db86e
	VCNT V13.B8, V11.B8                                         // ab59200e
	VMOV V31.B[15], V18                                         // f2071f5e
	VDUP V31.B[15], V18                                         // f2071f5e
	VDUP V31.B[13], V20.B16                                     // f4071b4e
	VEOR V4.B8, V18.B8, V7.B8                                   // 471e242e

func ARM64RegisterArrangement(reg int16, name, arng string) (int64, error) {
	var curQ, curSize uint16
	if name[0] != 'V' {
		return 0, errors.New("expect V0 through V31; found: " + name)
	}
	if reg < 0 {
		return 0, errors.New("invalid register number: " + name)
	}
	switch arng {
	case "B8":
		curSize = 0
		curQ = 0
	case "B16":
		curSize = 0
		curQ = 1

	VLD2R.P	(R0)(R5), [V31.D1, V0.D1]                       // 1fcce50d
	VLD3R	(RSP), [V31.S2, V0.S2, V1.S2]                   // ffeb400d
	VLD3R.P	6(R15), [V15.H4, V16.H4, V17.H4]                // efe5df0d
	VLD3R.P	(R15)(R6), [V15.H8, V16.H8, V17.H8]             // efe5c64d
	VLD4R	(R0), [V0.B8, V1.B8, V2.B8, V3.B8]              // 00e0600d
	VLD4R.P	16(RSP), [V31.S4, V0.S4, V1.S4, V2.S4]          // ffebff4d

INV V1 // The following macros are used for // the stitched implementation within // counterCryptASM. // Load the initial GCM counter value // in V30 and set up the counter increment // in V31 #define SETUP_COUNTER \ P8_LXVB16X(COUNTER, R0, V30); \ VSPLTISB $1, V28; \ VXOR V31, V31, V31; \ VSLDOI $1, V31, V28, V31 // These macros set up the initial value // for a single encryption, or 4 or 8 // stitched encryptions implemented // with interleaving vciphers. // // The input value for each encryption...

// registers in []. There may be comma-separated ranges or individual
// registers, as in [R1,R3-R5] or [V1.S4, V2.S4, V3.S4, V4.S4].
// For ARM, only R0 through R15 may appear.
// For ARM64, V0 through V31 with arrangement may appear.
//
// For 386/AMD64 register list specifies 4VNNIW-style multi-source operand.
// For range of 4 elements, Intel manual uses "+3" notation, for example:
//

	VMOVQ		V7.BU[0], R4    // e480f372
	VMOVQ		V7.BU[1], R4    // e484f372
	VMOVQ		V9.BU[3], R5    // 258df372
	VMOVQ		V10.HU[2], R6   // 46c9f372
	VMOVQ		V11.WU[2], R7   // 67e9f372
	VMOVQ		V31.VU[1], R8   // e8f7f372
	XVMOVQ		X1.W[2], R7     // 27c8ef76
	XVMOVQ		X6.V[2], R8     // c8e8ef76
	XVMOVQ		X8.WU[2], R7    // 07c9f376
	XVMOVQ		X31.VU[2], R8   // e8ebf376