words:
	// do 8 bytes at a time if there is room
	ADDV	$-7, R9, R6 // R6 is end pointer-7

	PCALIGN	$16
	SGTU	R6, R4, R8
	BEQ	R8, out
	MOVV	(R5), R7
	ADDV	$8, R5
	MOVV	R7, (R4)
	ADDV	$8, R4
	JMP	-6(PC)

out:
	BEQ	R4, R9, done
	MOVB	(R5), R7
	ADDV	$1, R5
	MOVB	R7, (R4)
	ADDV	$1, R4
	JMP	-5(PC)
done:
	RET

backward:
	ADDV	R6, R5 // from-end pointer
	ADDV	R4, R6, R9 // to-end pointer

TEXT ·IndexByte<ABIInternal>(SB),NOSPLIT,$0-40
	// R4 = b_base
	// R5 = b_len
	// R6 = b_cap (unused)
	// R7 = byte to find
	AND	$0xff, R7
	MOVV	R4, R6		// store base for later
	ADDV	R4, R5		// end
	ADDV	$-1, R4

	PCALIGN	$16
loop:
	ADDV	$1, R4
	BEQ	R4, R5, notfound
	MOVBU	(R4), R8
	BNE	R7, R8, loop

	SUBV	R6, R4		// remove base
	RET

notfound:
	MOVV	$-1, R4
	RET

TEXT runtime·memclrNoHeapPointers<ABIInternal>(SB),NOSPLIT,$0-16
	ADDV	R4, R5, R6

	// if less than 8 bytes, do one byte at a time
	SGTU	$8, R5, R8
	BNE	R8, out

	// do one byte at a time until 8-aligned
	AND	$7, R4, R8
	BEQ	R8, words
	MOVB	R0, (R4)
	ADDV	$1, R4
	JMP	-4(PC)

words:
	// do 8 bytes at a time if there is room
	ADDV	$-7, R6, R5

	PCALIGN	$16
	SGTU	R5, R4, R8
	BEQ	R8, out

	MOVV	32(R3), REGCTXT // restore REGCTXT

	MOVV	REGCTXT, 8(R3)
	MOVV	$argframe+0(FP), R20
	MOVV	R20, 16(R3)
	MOVV	R0, LOCAL_RETVALID(R3)
	ADDV	$LOCAL_RETVALID, R3, R20
	MOVV	R20, 24(R3)
	ADDV	$LOCAL_REGARGS, R3, R20
	MOVV	R20, 32(R3)
	JAL	·callReflect(SB)
	ADDV	$LOCAL_REGARGS, R3, R25	//unspillArgs using R25
	JAL	runtime·unspillArgs(SB)
	RET

#define	REGCTXT	R29

// memequal(a, b unsafe.Pointer, size uintptr) bool
TEXT runtime·memequal<ABIInternal>(SB),NOSPLIT|NOFRAME,$0-25
	BEQ	R4, R5, eq
	ADDV	R4, R6, R7
	PCALIGN	$16
loop:
	BNE	R4, R7, test
	MOVV	$1, R4
	RET
test:
	MOVBU	(R4), R9
	ADDV	$1, R4
	MOVBU	(R5), R10
	ADDV	$1, R5
	BEQ	R9, R10, loop

	MOVB    R0, R4
	RET
eq:
	MOVV	$1, R4
	RET

// memequal_varlen(a, b unsafe.Pointer) bool

	BNE	R0, R15, byte_loop

	PCALIGN	$16
chunk16_loop:
	BEQ	R0, R14, byte_loop
	MOVV	(R4), R8
	MOVV	(R6), R9
	BNE	R8, R9, byte_loop
	MOVV	8(R4), R16
	MOVV	8(R6), R17
	ADDV	$16, R4
	ADDV	$16, R6
	SUBVU	$1, R14
	BEQ	R16, R17, chunk16_loop
	SUBV	$8, R4
	SUBV	$8, R6

byte_loop:
	BEQ	R4, R12, samebytes
	MOVBU	(R4), R8
	ADDVU	$1, R4
	MOVBU	(R6), R9
	ADDVU	$1, R6
	BEQ	R8, R9, byte_loop

byte_cmp:

	fmt.Fprintln(buf, "BNE R5, label23")

	for i := 0; i <= genSize1; i++ {
		fmt.Fprintln(buf, "ADDV $0, R0, R0")
	}

	fmt.Fprintln(buf, "label18:")
	for i := 0; i <= (genSize2 - genSize1); i++ {
		fmt.Fprintln(buf, "ADDV $0, R0, R0")
	}

	fmt.Fprintln(buf, "label23:")
	fmt.Fprintln(buf, "ADDV $0, R0, R0")
	fmt.Fprintln(buf, "RET")

	// amd64:"NEGL","ADCQ","SBBQ","NEGQ"
	// loong64: "ADDV", "SGTU"
	// ppc64x: "ADDC", "ADDE", "ADDZE"
	// s390x:"ADDE","ADDC\t[$]-1,"
	// mips64:"ADDV","SGTU"
	// riscv64: "ADD","SLTU"
	return bits.Add(x, 7, ci)
}

func AddZ(x, y uint) (r, co uint) {
	// arm64:"ADDS","ADC",-"ADCS",-"ADD\t",-"CMP"
	// amd64:"ADDQ","SBBQ","NEGQ",-"NEGL",-"ADCQ"
	// loong64: "ADDV", "SGTU"
	// ppc64x: "ADDC", -"ADDE", "ADDZE"

	NO_LOCAL_POINTERS;			\
	/* copy arguments to stack */		\
	MOVV	arg+16(FP), R4;			\
	MOVWU	argsize+24(FP), R5;			\
	MOVV	R3, R12;				\
	ADDV	$8, R12;			\
	ADDV	R12, R5;				\
	BEQ	R12, R5, 6(PC);				\
	MOVBU	(R4), R6;			\
	ADDV	$1, R4;			\
	MOVBU	R6, (R12);			\
	ADDV	$1, R12;			\
	JMP	-5(PC);				\
	/* set up argument registers */		\
	MOVV	regArgs+40(FP), R25;		\
	JAL	·unspillArgs(SB);		\

	MOVV	R4, R5			// 85001500
	MOVBU	R4, R5			// 85fc4303
	SUB	R4, R5, R6		// a6101100
	SUBV	R4, R5, R6		// a6901100
	ADD	R4, R5, R6		// a6101000
	ADDV	R4, R5, R6		// a6901000
	AND	R4, R5, R6		// a6901400
	SUB	R4, R5			// a5101100
	SUBV	R4, R5			// a5901100
	ADD	R4, R5			// a5101000
	ADDV	R4, R5			// a5901000
	AND	R4, R5			// a5901400
	NEGW	R4, R5			// 05101100
	NEGV	R4, R5			// 05901100
	SLL	R4, R5			// a5101700