MOVD	$1(R7), R8
	CMPBGT	R8, R2, index2to16tail
index2to16loop:
	// unrolled 2x
	VLL	R4, (R7), V1
	VLL	R4, 1(R7), V2
	VCEQGS	V0, V1, V3
	BEQ	found
	MOVD	$1(R7), R7
	VCEQGS	V0, V2, V4
	BEQ	found
	MOVD	$1(R7), R7
	CMPBLT	R7, R2, index2to16loop
	CMPBGT	R7, R2, notfound
index2to16tail:
	VLL	R4, (R7), V1
	VCEQGS	V0, V1, V2
	BEQ	found
	BR	notfound

index17plus:

	// Jump to loop if we have more than 15 bytes to process.
	CGIJ $NE, R_ITER, $0, vxchunks

	// Load 1-15 bytes and corresponding mask.
	// Note: only the low 32-bits of R_LEN are used for the index.
	VLL R_LEN, (R_PTR), V_VAL
	VLL R_LEN, (R_MPTR), V_MASK

	// Compare each byte in input chunk against byte to be counted.
	// Each byte element will be set to either 0 (no match) or 1 (match).

	"VLLEZH",
	"VLLEZF",
	"VLLEZG",
	"VLM",
	"VLP",
	"VLPB",
	"VLPH",
	"VLPF",
	"VLPG",
	"VLBB",
	"VLVG",
	"VLVGB",
	"VLVGH",
	"VLVGF",
	"VLVGG",
	"VLVGP",
	"VLL",
	"VMX",
	"VMXB",
	"VMXH",
	"VMXF",
	"VMXG",
	"VMXL",
	"VMXLB",
	"VMXLH",
	"VMXLF",
	"VMXLG",
	"VMRH",
	"VMRHB",
	"VMRHH",
	"VMRHF",
	"VMRHG",
	"VMRL",
	"VMRLB",

	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
	VLEIF  $1, $1, INC
	VLEIF  $2, $2, INC
	VLEIF  $3, $3, INC

	CMPBLE R3, $16, b1

	// Load the 2 remaining blocks (17-32 bytes remaining).
	MOVD $-17(R3), R0    // index of final byte to load modulo 16
	VL   (R2), T_0       // load full 16 byte block
	VLL  R0, 16(R2), T_1 // load final (possibly partial) block and pad with zeros to 16 bytes

	// The Poly1305 algorithm requires that a 1 bit be appended to
	// each message block. If the final block is less than 16 bytes

000001f0  5b cf 99 b1 3e e6 f5 7b  1c d4 d0 d5 34 fc ab c6  |[...>..{....4...|
00000200  29 8e 0b 53 e4 7b 35 33  e4 e3 f6 77 9b f6 c6 ae  |)..S.{53...w....|
00000210  73 3c c5 80 c8 72 f7 56  6c 4c ce e5 a7 05 63 ee  |s<...r.VlL....c.|
00000220  07 e0 1d 3e 73 3e 55 73  ba 63 a8 d7 99 6a 48 57  |...>s>Us.c...jHW|
00000230  0c 55 51 42 4b 27 91 4b  75 b3 3b 4c 29 f1 52 e8  |.UQBK'.Ku.;L).R.|

	VGEG	$1, 8(R15)(V30*1), V31  // e7fef0081c12
	VSCEG	$1, V31, 16(R15)(V30*1) // e7fef0101c1a
	VGEF	$0, 2048(R15)(V1*1), V2 // e721f8000013
	VSCEF	$0, V2, 4095(R15)(V1*1) // e721ffff001b
	VLL	R0, (R15), V1           // e710f0000037
	VSTL	R0, V16, (R15)          // e700f000083f
	VGMH	$8, $16, V12            // e7c008101046
	VLEIB	$15, $255, V0           // e70000fff040
	VLEIH	$7, $-32768, V15        // e7f080007041