MOVQ		p_len+16(FP), numBytes		// get number of input bytes to hash
	SHRQ		$6, numBytes			// force modulo 64 input buffer length
	SHLQ		$6, numBytes
	CMPQ		numBytes, $0			// exit early for zero-length input buffer
	JEQ		done
	ADDQ		dataPtr, numBytes		// point numBytes to end of input buffer
	VMOVDQU		(0*16)(digestPtr), state0	// load initial hash values and reorder
	VMOVDQU		(1*16)(digestPtr), state1	// DCBA, HGFE -> ABEF, CDGH

	MOVHZ	(MASK_PTR)(R16), R14
	XOR	R14, R15, R14
	MOVH	R14, (R16)(BLK_OUT)
	ADD	$2, R16
	ADD	$-2, IN_LEN
next1:
	CMP	IN_LEN, $1
	BLT	done
	MOVBZ	(MASK_PTR)(R16), R14
	MOVBZ	(BLK_INP)(R16), R15
	XOR	R14, R15, R14
	MOVB	R14, (R16)(BLK_OUT)
#endif
done:
	// Save the updated counter value
	P8_STXVB16X(V30, COUNTER, R0)
	// Clear the keys
	XXLXOR	VS0, VS0, VS0
	XXLXOR	VS1, VS1, VS1

		mulRound(B3)
		mulRound(B4)
		mulRound(B5)
		mulRound(B6)
		mulRound(B7)
		MOVD	pTblSave, pTbl
		reduce()

		CMP	$128, srcPtrLen
		BGE	octetsLoop

startSingles:
	CBZ	srcPtrLen, done
	ADD	$14*16, pTbl
	// Preload H and its Karatsuba precomp
	VLD1.P	(pTbl), [T1.B16, T2.B16]
	// Preload AES round keys
	ADD	$128, ks
	VLD1.P	48(ks), [K8.B16, K9.B16, K10.B16]
	VMOV	K10.B16, KLAST.B16

	VXOR	IN0, KEY, IN0                    // vxor 1,1,3
	STXVD2X	IN0, (R0+OUTENC)
	STXVD2X	IN0, (R0+OUTDEC)
	ADD	$16, OUTENC, OUTENC
	ADD	$-16, OUTDEC, OUTDEC
	BDZ	done

	VSPLTW	$3, IN0, KEY        // vspltw 3,1,3
	VSLDOI	$12, ZERO, IN1, TMP // vsldoi 6,0,2,12
	VSBOX	KEY, KEY            // vsbox 3,3

	VXOR	IN1, TMP, IN1       // vxor 2,2,6