MOVD R9, R6
	MOVD R2, R7

	// n >= 0
	// regular loop body unrolled 16x
	VZERO V0 // c = 0

UU1:
	VLM  0(R5), V1, V4    // 64-bytes into V1..V8
	ADD  $64, R5
	VPDI $0x4, V1, V1, V1 // flip the doublewords to big-endian order
	VPDI $0x4, V2, V2, V2 // flip the doublewords to big-endian order

	VLM  0(R6), V9, V12      // 64-bytes into V9..V16
	ADD  $64, R6
	VPDI $0x4, V9, V9, V9    // flip the doublewords to big-endian order

	MOVD key+48(FP), R5        // R5=key
	MOVD nonce+56(FP), R6      // R6=nonce
	MOVD counter+64(FP), R7    // R7=counter

	// load BSWAP and J0
	VLM (R1), BSWAP, J0

	// setup
	MOVD  $95, R0
	VLM   (R5), KEY0, KEY1
	VLL   R0, (R6), NONCE
	VZERO M0
	VLEIB $7, $32, M0
	VSRLB M0, NONCE, NONCE

	// initialize counter values
	VLREPF (R7), CTR
	VZERO  INC

TEXT vectorizedBody<>(SB),NOSPLIT,$0
	XOR     $0xffffffff, R2 // NOTW R2
	VLM     0(R5), CONST_PERM_LE2BE, CONST_CRC_POLY

	// Load the initial CRC value into the rightmost word of V0
	VZERO   V0
	VLVGF   $3, R2, V0

	// Crash if the input size is less than 64-bytes.
	CMP     R4, $64
	BLT     crash

	// Load a 64-byte data chunk and XOR with CRC
	VLM     0(R3), V1, V4    // 64-bytes into V1..V4

	"VLEIF",
	"VLEIG",
	"VLEIB",
	"VFI",
	"VFIDB",
	"WFIDB",
	"VLGV",
	"VLGVB",
	"VLGVH",
	"VLGVF",
	"VLGVG",
	"VLLEZ",
	"VLLEZB",
	"VLLEZH",
	"VLLEZF",
	"VLLEZG",
	"VLM",
	"VLP",
	"VLPB",
	"VLPH",
	"VLPF",
	"VLPG",
	"VLBB",
	"VLVG",
	"VLVGB",
	"VLVGH",
	"VLVGF",
	"VLVGG",
	"VLVGP",
	"VLL",
	"VMX",
	"VMXB",
	"VMXH",
	"VMXF",

// func updateVX(state *macState, msg []byte)
TEXT ·updateVX(SB), NOSPLIT, $0
	MOVD state+0(FP), R1
	LMG  msg+8(FP), R2, R3 // R2=msg_base, R3=msg_len

	// load EX0, EX1 and EX2
	MOVD $·constants<>(SB), R5
	VLM  (R5), EX0, EX2

	// generate masks
	VGMG $(64-24), $63, MOD24 // [0x00ffffff, 0x00ffffff]
	VGMG $(64-26), $63, MOD26 // [0x03ffffff, 0x03ffffff]

	// load h (accumulator) and r (key) from state

	VL	(R15), V31              // e7f0f0000806
	VST	V31, (R15)              // e7f0f000080e
	VESLB	$5, V14                 // e7ee00050030
	VESRAG	$0, V15, V16            // e70f0000383a
	VLM	(R15), V8, V23          // e787f0000436
	VSTM	V8, V23, (R15)          // e787f000043e
	VONE	V1                      // e710ffff0044
	VZERO	V16                     // e70000000844
	VGBM	$52428, V31             // e7f0cccc0844