XXLXOR VS32, VS32, VS32 // clear VS32 (V0)
	BEQ   lt64gt8

	// Prepare to clear 64 bytes at a time.

zero64setup:
	DCBTST (R3)             // prepare data cache
	MOVD   R7, CTR          // number of 64 byte chunks
	MOVD   $16, R8
	MOVD   $32, R16
	MOVD   $48, R17

zero64:
	STXVD2X VS32, (R3+R0)   // store 16 bytes
	STXVD2X VS32, (R3+R8)
	STXVD2X VS32, (R3+R16)
	STXVD2X VS32, (R3+R17)
	ADD     $64, R3

	CMP	DWORDS, $2

	MOVD 	-8(SRC), TMP
	MOVD 	TMP, -8(TGT)
	BC	12, 0, LR		// bltlr, return if DWORDS == 1

	MOVD 	-16(SRC), TMP
	MOVD 	TMP, -16(TGT)
	BC	12, 2, LR		// beqlr, return if DWORDS == 2

	MOVD 	-24(SRC), TMP
	MOVD 	TMP, -24(TGT)
	RET

backward32setup:
	ANDCC   $3,DWORDS		// Compute remaining DWORDS and compare to 0
	MOVD	QWORDS, CTR		// set up loop ctr

	MOVWZ	ctab+4(FP), R14   // crc poly id
	MOVD    p+8(FP), R4
	MOVD    p_len+16(FP), R5 // p len

	// R3 = incoming crc
	// R14 = constant table identifier
	// R5 = address of bytes
	// R6 = length of bytes

	// defines for index loads

	MOVD	$16,off16
	MOVD	$32,off32
	MOVD	$48,off48
	MOVD	$64,off64
	MOVD	$80,off80
	MOVD	$96,off96
	MOVD	$112,off112
	MOVD	$0,R15

	BLT	tail
	// the length of tail > 8 bytes
	MOVD.P	8(R0), R4
	MOVD.P	8(R2), R5
	CMP	R4, R5
	BNE	cmp
	SUB	$8, R6
	// compare last 8 bytes
tail:
	MOVD	(R0)(R6), R4
	MOVD	(R2)(R6), R5
	CMP	R4, R5
	BEQ	samebytes
cmp:
	REV	R4, R4
	REV	R5, R5
	CMP	R4, R5
ret:
	MOVD	$1, R0
	CNEG	HI, R0, R0
	RET
small:
	TBZ	$3, R6, lt_8
	MOVD	(R0), R4
	MOVD	(R2), R5
	CMP	R4, R5
	BNE	cmp
	SUBS	$8, R6

// For keyp of 12, V6, V9-V20 hold the expanded key.
// For keyp of 14, V6, V7-V20 hold the expanded key.
#define LOAD_KEY(Rkeyp) \
	MOVD	$16, R12 \
	MOVD	$32, R14 \
	MOVD	$48, R15 \
	MOVD	$64, R16 \
	MOVD	$80, R17 \
	MOVD	$96, R18 \
	MOVD	$112, R19 \
	MOVD	$128, R20 \
	MOVD	$144, R21 \
	LXVD2X	(R0+Rkeyp), V6 \
	ADD	$16, Rkeyp \
	BEQ	CR1, L_start10 \
	BEQ	CR2, L_start12 \
	LXVD2X	(R0+Rkeyp), V7 \

#include "textflag.h"

TEXT ·IndexByte(SB),NOSPLIT,$0-40
	MOVD	b_base+0(FP), R0
	MOVD	b_len+8(FP), R2
	MOVBU	c+24(FP), R1
	MOVD	$ret+32(FP), R8
	B	indexbytebody<>(SB)

TEXT ·IndexByteString(SB),NOSPLIT,$0-32
	MOVD	s_base+0(FP), R0
	MOVD	s_len+8(FP), R2
	MOVBU	c+16(FP), R1
	MOVD	$ret+24(FP), R8
	B	indexbytebody<>(SB)

// input:
//   R0: data
//   R1: byte to search

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

//go:build !purego

#include "textflag.h"

// func xorBytes(dst, a, b *byte, n int)
TEXT ·xorBytes(SB), NOSPLIT|NOFRAME, $0
	MOVD	dst+0(FP), R0
	MOVD	a+8(FP), R1
	MOVD	b+16(FP), R2
	MOVD	n+24(FP), R3
	CMP	$64, R3
	BLT	tail
loop_64:
	VLD1.P	64(R1), [V0.B16, V1.B16, V2.B16, V3.B16]
	VLD1.P	64(R2), [V4.B16, V5.B16, V6.B16, V7.B16]
	VEOR	V0.B16, V4.B16, V4.B16

	// arm64:`MOVD\t\$12592`,`MOVD\t\$50`
	//  wasm:`I64Const\t\$12592`,`I64Store16\t\$0`,`I64Const\t\$50`,`I64Store8\t\$2`
	// mips64:`MOVV\t\$48`,`MOVV\t\$49`,`MOVV\t\$50`
	bsink = []byte("012")

	// 858927408 = 0x33323130
	//     13620 = 0x3534
	// amd64:`MOVL\t\$858927408`,`MOVW\t\$13620, 4\(`
	//   386:`MOVL\t\$858927408`,`MOVW\t\$13620, 4\(`
	// arm64:`MOVD\t\$858927408`,`MOVD\t\$13620`

	CMP	R0, R6
	BNE	chunk16_loop
	AND	$0xf, R2, R2
	CBZ	R2, equal
tail:
	// special compare of tail with length < 16
	TBZ	$3, R2, lt_8
	MOVD	(R0), R4
	MOVD	(R1), R5
	EOR	R4, R5
	CBNZ	R5, not_equal
	SUB	$8, R2, R6	// offset of the last 8 bytes
	MOVD	(R0)(R6), R4
	MOVD	(R1)(R6), R5
	EOR	R4, R5
	CBNZ	R5, not_equal
	B	equal
lt_8:
	TBZ	$2, R2, lt_4
	MOVWU	(R0), R4
	MOVWU	(R1), R5
	EOR	R4, R5

	// s390x:`MOVDBR\s\(.*\),`
	// arm64:`MOVD\s\(R[0-9]+\),`,-`MOV[BHW]`
	// ppc64le:`MOVD\s`,-`MOV[BHW]Z`
	// ppc64:`MOVDBR\s`,-`MOV[BHW]Z`
	return binary.LittleEndian.Uint64(b)
}

func load_le64_idx(b []byte, idx int) uint64 {
	// amd64:`MOVQ\s\(.*\)\(.*\*1\),`,-`MOV[BWL]\t[^$]`,-`OR`
	// s390x:`MOVDBR\s\(.*\)\(.*\*1\),`
	// arm64:`MOVD\s\(R[0-9]+\)\(R[0-9]+\),`,-`MOV[BHW]`
	// ppc64le:`MOVD\s`,-`MOV[BHW]Z\s`