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]
	RLDICL	$48,R7,$56,R24	// crc>>16&0xFF

	XOR	R0, R0
	CMP	R22, $0
	BEQ	good
	MOVD	48(R1), R22
	MOVD	$8, R20
	TW	$31, R0, R0

	BR	restore

good:
#define DEBUG_CALL_DISPATCH(NAME,MAXSIZE)	\
	MOVD	$MAXSIZE, R23;			\
	CMP	R26, R23;			\
	BGT	5(PC);				\
	MOVD	$NAME(SB), R26;			\
	MOVD	R26, 32(R1);			\
	CALL	runtime·debugCallWrap(SB);	\
	BR	restore

	// the argument frame size
	MOVD	128(R1), R26

#define M02 R20
#define M03 R24
#define M04 R25
#define M05 R26
#define M06 R27
#define M07 R28
#define M08 R29
#define M09 R21
#define M10 R11
#define M11 R8
#define M12 R7
#define M13 R12
#define M14 R23
#define M15 R10

#define ROUND1(a, b, c, d, index, const, shift) \
	ADD	$const, index, R9; \
	ADD	R9, a; \
	AND     b, c, R9; \
	ANDN    b, d, R31; \
	OR	R9, R31, R9; \
	ADD	R9, a; \
	ROTLW	$shift, a; \

	}
	fmt.Fprintln(w, "\tRET")
}

func copyMIPS64x(w io.Writer) {
	fmt.Fprintln(w, "TEXT runtime·duffcopy(SB), NOSPLIT|NOFRAME, $0-0")
	for i := 0; i < 128; i++ {
		fmt.Fprintln(w, "\tMOVV\t(R1), R23")
		fmt.Fprintln(w, "\tADDV\t$8, R1")
		fmt.Fprintln(w, "\tMOVV\tR23, (R2)")
		fmt.Fprintln(w, "\tADDV\t$8, R2")
		fmt.Fprintln(w)
	}
	fmt.Fprintln(w, "\tRET")
}

func zeroRISCV64(w io.Writer) {

	STP	(R12, R13), 11*8(RSP)
	STP	(R14, R15), 13*8(RSP)
	// R16, R17 may be clobbered by linker trampoline
	// R18 is unused.
	STP	(R19, R20), 15*8(RSP)
	STP	(R21, R22), 17*8(RSP)
	STP	(R23, R24), 19*8(RSP)
	STP	(R25, R26), 21*8(RSP)
	// R27 is temp register.
	// R28 is g.
	// R29 is frame pointer (unused).
	// R30 is LR, which was saved by the prologue.
	// R31 is SP.

#define R_x030	R26
#define R_x040	R14
#define R_x050	R15
#define R_x060	R16
#define R_x070	R17
#define R_x080	R18
#define R_x090	R19
#define R_x0a0	R20
#define R_x0b0	R21
#define R_x0c0	R22
#define R_x0d0	R23
#define R_x0e0	R24
#define R_x0f0	R28
#define R_x100	R29
#define R_x110	R27


// V0-V7 are A-H
// V8-V23 are used for the message schedule
#define KI	V24
#define FUNC	V25
#define S0	V26

#define R_x030	R11
#define R_x040	R14
#define R_x050	R15
#define R_x060	R16
#define R_x070	R17
#define R_x080	R18
#define R_x090	R19
#define R_x0a0	R20
#define R_x0b0	R21
#define R_x0c0	R22
#define R_x0d0	R23
#define R_x0e0	R24
#define R_x0f0	R25
#define R_x100	R26
#define R_x110	R27


// V0-V7 are A-H
// V8-V23 are used for the message schedule
#define KI	V24
#define FUNC	V25
#define S0	V26

#define R_x030	R11
#define R_x040	R14
#define R_x050	R15
#define R_x060	R16
#define R_x070	R17
#define R_x080	R18
#define R_x090	R19
#define R_x0a0	R20
#define R_x0b0	R21
#define R_x0c0	R22
#define R_x0d0	R23
#define R_x0e0	R24
#define R_x0f0	R25
#define R_x100	R26
#define R_x110	R27


// V0-V7 are A-H
// V8-V23 are used for the message schedule
#define KI	V24
#define FUNC	V25
#define S0	V26

	&& !config.noDuffDevice && logLargeCopy(v, s)  =>
	(DUFFCOPY [16 * (128 - s/8)] dst src mem)
// 16 and 128 are magic constants.  16 is the number of bytes to encode:
//	MOVV	(R1), R23
//	ADDV	$8, R1
//	MOVV	R23, (R2)
//	ADDV	$8, R2
// and 128 is the number of such blocks. See runtime/duff_mips64.s:duffcopy.

// large or unaligned move uses a loop
(Move [s] {t} dst src mem)

				// if(g->panic != nil && g->panic->argp == FP) g->panic->argp = bottom-of-frame
				//
				//	MOVD g_panic(g), R22
				//	CMP R22, $0
				//	BEQ end
				//	MOVD panic_argp(R22), R23
				//	ADD $(autosize+8), R1, R24
				//	CMP R23, R24
				//	BNE end
				//	ADD $8, R1, R25
				//	MOVD R25, panic_argp(R22)
				// end:
				//	NOP
				//
				// The NOP is needed to give the jumps somewhere to land.