// bytes.
	//
	// CRC(I, ABC) = CRC(I, ABO xor C)
	//             = CRC(I, ABO) xor CRC(0, C)
	//             = CRC(CRC(I, AB), O) xor CRC(0, C)
	//             = CRC(CRC(I, AO xor B), O) xor CRC(0, C)
	//             = CRC(CRC(I, AO) xor CRC(0, B), O) xor CRC(0, C)
	//             = CRC(CRC(CRC(I, A), O) xor CRC(0, B), O) xor CRC(0, C)
	//
	// The castagnoliSSE42Triple function can compute CRC(I, A), CRC(0, B),

// vectorizedCastagnoli.
func archUpdateCastagnoli(crc uint32, p []byte) uint32 {
	if !hasVX {
		panic("not available")
	}
	// Use vectorized function if data length is above threshold.
	if len(p) >= vxMinLen {
		aligned := len(p) & ^vxAlignMask
		crc = vectorizedCastagnoli(crc, p[:aligned])
		p = p[aligned:]
	}
	if len(p) == 0 {
		return crc
	}
	return slicingUpdate(crc, archCastagnoliTable8, p)
}

	{0x336776c420000000, 0x3c9d28596e5960ba, "abcd", "crc\x02s\xba\x84\x84\xbb\xcd]\xef6\xc4 \x00\x00\x00\x00\x00", "crc\x02`&\x9aR\xe1\xb7\xfee\xbces \x0e\x84\xb0F"},
	{0x32d36776c4200000, 0x40bdf58fb0895f2, "abcde", "crc\x02s\xba\x84\x84\xbb\xcd]\xef6\xc4 \x00\x00\x00\x00\x00", "crc\x02`&\x9aR\xe1\xb7\xfee\xbces \x0e\x84\xb0F"},

	MOVWZ	(R10)(R8),R25	// tab[4][crc>>24]
	RLDICL	$48,R7,$56,R24	// crc>>16&0xFF
	XOR	R21,R25,R21	// xor done R25
	ADD	$1024,R10,R10	// &tab[5]
	SLD	$2,R24,R24	// crc>>16&0xFF*4
	MOVWZ	(R10)(R24),R26	// tab[5][crc>>16&0xFF]
	XOR	R21,R26,R21	// xor done R26
	RLDICL	$56,R7,$56,R25	// crc>>8
	ADD	$1024,R10,R10	// &tab[6]
	SLD	$2,R25,R25	// crc>>8&FF*2
	MOVBZ   R7,R26          // crc&0xFF
	MOVWZ	(R10)(R25),R27	// tab[6][crc>>8&0xFF]

	{0x4b8e39ef, 0x53bceff1, "abcdef", "crc\x01ʇ\x91M5$A\xc2", "crc\x01wB\x84\x816K?\xb7"},
	{0x312a6aa6, 0xe627f441, "abcdefg", "crc\x01ʇ\x91M5$A\xc2", "crc\x01wB\x84\x816K?\xb7"},
	{0xaeef2a50, 0xa9421b7, "abcdefgh", "crc\x01ʇ\x91M\xed\x82\xcd\x11", "crc\x01wB\x84\x81\x92\xc8\n1"},
	{0x8da988af, 0x2ddc99fc, "abcdefghi", "crc\x01ʇ\x91M\xed\x82\xcd\x11", "crc\x01wB\x84\x81\x92\xc8\n1"},

// Register usage:
//
//      R5:     CRC-32 constant pool base pointer.
//      V0:     Initial CRC value and intermediate constants and results.
//      V1..V4: Data for CRC computation.
//      V5..V8: Next data chunks that are fetched from the input buffer.
//
//      V9..V14: CRC-32 constants.

// func vectorizedIEEE(crc uint32, p []byte) uint32
TEXT ·vectorizedIEEE(SB),NOSPLIT,$0
	MOVWZ   crc+0(FP), R2     // R2 stores the CRC value

		aligned := len(p) & ^vecAlignMask
		crc = vectorCrc32(crc, crcCast, p[:aligned])
		p = p[aligned:]
	}
	if len(p) == 0 {
		return crc
	}
	return ppc64SlicingUpdateBy8(crc, archCastagnoliTable8, p)
}

func archAvailableIEEE() bool {
	return true
}
func archAvailableCastagnoli() bool {
	return true
}

var archIeeeTable8 *slicing8Table

func archInitIEEE() {

	}
	d.crc = byteorder.BeUint32(b[8:])
	return nil
}

func update(crc uint32, tab *Table, p []byte, checkInitIEEE bool) uint32 {
	switch {
	case haveCastagnoli.Load() && tab == castagnoliTable:
		return updateCastagnoli(crc, p)
	case tab == IEEETable:
		if checkInitIEEE {
			ieeeOnce.Do(ieeeInit)
		}
		return updateIEEE(crc, p)
	default:
		return simpleUpdate(crc, tab, p)

package crc32

import "internal/cpu"

func castagnoliUpdate(crc uint32, p []byte) uint32
func ieeeUpdate(crc uint32, p []byte) uint32

func archAvailableCastagnoli() bool {
	return cpu.ARM64.HasCRC32
}

func archInitCastagnoli() {
	if !cpu.ARM64.HasCRC32 {
		panic("arch-specific crc32 instruction for Castagnoli not available")
	}
}

func archUpdateCastagnoli(crc uint32, p []byte) uint32 {
	if !cpu.ARM64.HasCRC32 {

// license that can be found in the LICENSE file.

#include "textflag.h"

// castagnoliUpdate updates the non-inverted crc with the given data.

// func castagnoliUpdate(crc uint32, p []byte) uint32
TEXT ·castagnoliUpdate(SB),NOSPLIT,$0-36
	MOVWU	crc+0(FP), R9  // CRC value
	MOVD	p+8(FP), R13  // data pointer
	MOVD	p_len+16(FP), R11  // len(p)

update:
	CMP	$16, R11
	BLT	less_than_16