AND	$0xff, X13, X12
	MOV	ZERO, X14	// count
	ADD	X10, X11	// end

	PCALIGN	$16
loop:
	BEQ	X10, X11, done
	MOVBU	(X10), X15
	ADD	$1, X10
	BNE	X12, X15, loop
	ADD	$1, X14
	JMP	loop

done:
	MOV	X14, X10
	RET

TEXT ·CountString<ABIInternal>(SB),NOSPLIT,$0-32
	// X10 = s_base
	// X11 = s_len
	// X12 = byte to count
	AND	$0xff, X12
	MOV	ZERO, X14	// count
	ADD	X10, X11	// end

				p.Spadj = int32(-p.From.Offset)
			}

		case obj.AGETCALLERPC:
			if cursym.Leaf() {
				// MOV LR, Rd
				p.As = mov
				p.From.Type = obj.TYPE_REG
				p.From.Reg = REGLINK
			} else {
				// MOV (RSP), Rd
				p.As = mov
				p.From.Type = obj.TYPE_MEM
				p.From.Reg = REGSP
			}
		}

		if p.To.Type == obj.TYPE_REG && p.To.Reg == REGSP && p.Spadj == 0 {

	fmt.Fprintln(w, "TEXT runtime·duffcopy<ABIInternal>(SB), NOSPLIT|NOFRAME, $0-0")
	for i := 0; i < 128; i++ {
		fmt.Fprintln(w, "\tMOV\t(X24), X31")
		fmt.Fprintln(w, "\tADD\t$8, X24")
		fmt.Fprintln(w, "\tMOV\tX31, (X25)")
		fmt.Fprintln(w, "\tADD\t$8, X25")
		fmt.Fprintln(w)
	}
	fmt.Fprintln(w, "\tRET")

	buf.WriteString(`// Code generated by wincallback.go using 'go generate'. DO NOT EDIT.

// External code calls into callbackasm at an offset corresponding
// to the callback index. Callbackasm is a table of MOV and B instructions.
// The MOV instruction loads R12 with the callback index, and the
// B instruction branches to callbackasm1.
// callbackasm1 takes the callback index from R12 and

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

#include "textflag.h"

TEXT _rt0_riscv64_openbsd(SB),NOSPLIT|NOFRAME,$0
	MOV	0(X2), A0	// argc
	ADD	$8, X2, A1	// argv
	JMP	main(SB)

TEXT main(SB),NOSPLIT|NOFRAME,$0
	MOV	$runtime·rt0_go(SB), T0

				p.Spadj = int32(-p.From.Offset)
			}

		case obj.AGETCALLERPC:
			if cursym.Leaf() {
				/* MOV LR, Rd */
				p.As = mov
				p.From.Type = obj.TYPE_REG
				p.From.Reg = REGLINK
			} else {
				/* MOV (RSP), Rd */
				p.As = mov
				p.From.Type = obj.TYPE_MEM
				p.From.Reg = REGSP
			}
		}

		if p.To.Type == obj.TYPE_REG && p.To.Reg == REGSP && p.Spadj == 0 {

	CRnEQ represents CR bit 2 of CR field n. (0-7)
	CRnSO represents CR bit 3 of CR field n. (0-7)

# GOPPC64 >= power10 and its effects on Go asm

When GOPPC64=power10 is used to compile a Go program for ppc64le/linux, MOV*, FMOV*, and ADD
opcodes which would require 2 or more machine instructions to emulate a 32 bit constant, or
symbolic reference are implemented using prefixed instructions.

	// amd64:"(MOVB\t[$]0)|(XORL\t[A-Z][A-Z0-9]+, [A-Z][A-Z0-9]+)",-"FCMP",-"MOVB\t[$]1"
	// s390x:"MOV(B|BZ|D)\t[$]0,",-"FCMPU",-"MOV(B|BZ|D)\t[$]1,"
	return 0.5 == float64(uint32(1)) || 1.5 > float64(uint64(1<<63))
}

func constantCheck32() bool {
	// amd64:"MOV(B|L)\t[$]1",-"FCMP",-"MOV(B|L)\t[$]0"
	// s390x:"MOV(B|BZ|D)\t[$]1,",-"FCMPU",-"MOV(B|BZ|D)\t[$]0,"
	return float32(0.5) <= float32(int64(1)) && float32(1.5) >= float32(int32(-1<<31))

// For example:
// (CMPBconstload c (ADDQ x y)) -> (CMPBconstloadidx1 c x y) -> (CMPB c (MOVBloadidx1 x y))

(CMP(Q|L|W|B)load {sym} [off] ptr x mem) => (CMP(Q|L|W|B) (MOV(Q|L|W|B)load {sym} [off] ptr mem) x)

(CMP(Q|L|W|B)constload {sym} [vo] ptr mem) && vo.Val() == 0 => (TEST(Q|L|W|B) x:(MOV(Q|L|W|B)load {sym} [vo.Off()] ptr mem) x)

(CMPQconstload {sym} [vo] ptr mem) && vo.Val() != 0 => (CMPQconst (MOVQload {sym} [vo.Off()] ptr mem) [vo.Val()])

TEXT ·RewindAndSetgid(SB),NOSPLIT|NOFRAME,$0-0
	// Rewind stack pointer so anything that happens on the stack
	// will clobber the test pattern created by the caller
	ADD	$(1024*8), X2

	// Ask signaller to setgid
	MOV	$1, X5
	FENCE
	MOVW	X5, ·Baton(SB)
	FENCE

	// Wait for setgid completion
loop:
	FENCE
	MOVW	·Baton(SB), X5
	OR	X6, X6, X6	// hint that we're in a spin loop
	BNE	ZERO, X5, loop
	FENCE