SLD	$2,R19,R19	// p[5]*4:1
	MOVWZ	(R10)(R18),R22	// tab[1][p[6]]
	ADD	$1024,R10,R10	// tab[2]
	XOR	R21,R22,R21	// xor done R22
	CLRLSLDI $56,R9,$2,R20
	MOVWZ	(R10)(R19),R23	// tab[2][p[5]]
	ADD	$1024,R10,R10	// &tab[3]
	XOR	R21,R23,R21	// xor done R23
	MOVWZ	(R10)(R20),R24	// tab[3][p[4]]
	ADD 	$1024,R10,R10   // &tab[4]
	XOR	R21,R24,R21	// xor done R24
	MOVWZ	(R10)(R8),R25	// tab[4][crc>>24]

	ROUND1(R3,R4,R5,R2, 0,0x49b40821,22);

	MOVWBR	(1*4)(R6), R8
	MOVWZ	R5, R9
	MOVWZ	R5, R1

#define ROUND2(a, b, c, d, index, const, shift) \
	XOR	$0xffffffff, R9; \ // NOTW R9
	ADD	$const, a; \
	ADD	R8, a; \
	MOVWBR	(index*4)(R6), R8; \
	AND	b, R1; \
	AND	c, R9; \
	OR	R9, R1; \
	MOVWZ	c, R9; \
	ADD	R1, a; \
	MOVWZ	c, R1; \
	RLL	$shift,	a; \
	ADD	b, a

	ROUND2(R2,R3,R4,R5, 6,0xf61e2562, 5);

	ADD	b, a;


TEXT ·block(SB),NOSPLIT,$0-32
	MOVD	dig+0(FP), R10
	MOVD	p+8(FP), R6
	MOVD	p_len+16(FP), R5

	// We assume p_len >= 64
	SRD 	$6, R5
	MOVD	R5, CTR

	MOVWZ	0(R10), R22
	MOVWZ	4(R10), R3
	MOVWZ	8(R10), R4
	MOVWZ	12(R10), R5

loop:
	MOVD	R22, R14
	MOVD	R3, R15
	MOVD	R4, R16
	MOVD	R5, R17

	ENDIAN_MOVE( 0,R6,M00,M15)
	ENDIAN_MOVE( 4,R6,M01,M15)
	ENDIAN_MOVE( 8,R6,M02,M15)

// bool cas(uint32 *ptr, uint32 old, uint32 new)
// Atomically:
//	if(*val == old){
//		*val = new;
//		return 1;
//	} else
//		return 0;
TEXT ·Cas(SB), NOSPLIT, $0-17
	MOVD	ptr+0(FP), R3
	MOVWZ	old+8(FP), R4
	MOVWZ	new+12(FP), R5
	LWSYNC
cas_again:
	LWAR	(R3), R6
	CMPW	R6, R4
	BNE	cas_fail
	STWCCC	R5, (R3)
	BNE	cas_again
	MOVD	$1, R3
	LWSYNC
	MOVB	R3, ret+16(FP)
	RET
cas_fail:
	LWSYNC

	// ppc64x:-"MOVW\tR\\d+,\\sR\\d+"
	sval64[2] = int64(*x32)

	// ppc64x:-"MOVBZ\tR\\d+,\\sR\\d+"
	val64[0] = uint64(*u8)

	// ppc64x:-"MOVHZ\tR\\d+,\\sR\\d+"
	val64[1] = uint64(*u16)

	// ppc64x:-"MOVWZ\tR\\d+,\\sR\\d+"
	val64[2] = uint64(*u32)
}

func cmp16(u8 *uint8, x32 *int32, u32 *uint32, x64 *int64, u64 *uint64) bool {

	// ppc64x:-"MOVBZ\tR\\d+,\\sR\\d+"
	if uint16(*u8) == val16[0] {
		return true
	}

	XOR	R14, R15, R16   // R16 = a[] ^ b[]
	SUB	$8, R6          // n = n - 8
	MOVD	R16, (R3)(R8)   // Store to dst
	ADD	$8, R8
xor4:
	CMP	R6, $4
	BLT	xor2
	MOVWZ	(R4)(R8), R14
	MOVWZ	(R5)(R8), R15
	XOR	R14, R15, R16
	MOVW	R16, (R3)(R8)
	ADD	$4,R8
	ADD	$-4,R6
xor2:
	CMP	R6, $2
	BLT	xor1
	MOVHZ	(R4)(R8), R14
	MOVHZ	(R5)(R8), R15
	XOR	R14, R15, R16

	MOVD	p_len+16(FP), LEN

	SRD	$6, LEN
	SLD	$6, LEN

	ADD	INP, LEN, END

	CMP	INP, END
	BEQ	end

	MOVD	$·kcon(SB), TBL_STRT

	MOVD	R0, CNT
	MOVWZ	$0x010, R_x010
	MOVWZ	$0x020, R_x020
	MOVWZ	$0x030, R_x030
	MOVD	$0x040, R_x040
	MOVD	$0x050, R_x050
	MOVD	$0x060, R_x060
	MOVD	$0x070, R_x070
	MOVD	$0x080, R_x080
	MOVD	$0x090, R_x090
	MOVD	$0x0a0, R_x0a0

	MOVD  bpx_offset+24(FP), R2 // r2 offset to BPX vector table
	MOVD  R14, R7               // save r14
	MOVD  R15, R8               // save r15
	MOVWZ 16(R0), R9
	MOVWZ 544(R9), R9
	MOVWZ 24(R9), R9            // call vector in r9
	ADD   R2, R9                // add offset to vector table
	MOVWZ (R9), R9              // r9 points to entry point
	BYTE  $0x0D                 // BL R14,R9 --> basr r14,r9

// func Cas(ptr *uint32, old, new uint32) bool
// Atomically:
//	if *ptr == old {
//		*val = new
//		return 1
//	} else {
//		return 0
//	}
TEXT ·Cas(SB), NOSPLIT, $0-17
	MOVD	ptr+0(FP), R3
	MOVWZ	old+8(FP), R4
	MOVWZ	new+12(FP), R5
	CS	R4, R5, 0(R3)    //  if (R4 == 0(R3)) then 0(R3)= R5
	BNE	cas_fail
	MOVB	$1, ret+16(FP)
	RET
cas_fail:
	MOVB	$0, ret+16(FP)
	RET

// func Cas64(ptr *uint64, old, new uint64) bool

	// RX (12-bit displacement) and RXY (20-bit displacement) instruction encoding (e.g: ST vs STY)
	MOVW	R1, 4095(R2)(R3)       // 50132fff
	MOVW	R1, 4096(R2)(R3)       // e31320000150
	MOVWZ	R1, 4095(R2)(R3)       // 50132fff
	MOVWZ	R1, 4096(R2)(R3)       // e31320000150
	MOVH	R1, 4095(R2)(R3)       // 40132fff
	MOVHZ   R1, 4095(R2)(R3)       // 40132fff
	MOVH	R1, 4096(R2)(R3)       // e31320000170