nodes.Append(nif)

	// Index to start copying into s.
	//   idx = newLen - len(l2)
	// We use this expression instead of oldLen because it avoids
	// a spill/restore of oldLen.
	// Note: this doesn't work optimally currently because
	// the compiler optimizer undoes this arithmetic.
	idx := ir.NewBinaryExpr(base.Pos, ir.OSUB, newLen, ir.NewUnaryExpr(base.Pos, ir.OLEN, l2))

		//	STP.P	(ZR,ZR), 16(R16)
		//	CMP	Rarg1, R16
		//	BLE	-2(PC)
		// Note: the-end-of-the-memory may be not a valid pointer. it's a problem if it is spilled.
		// the-end-of-the-memory - 16 is with the area to zero, ok to spill.
		{
			name:      "LoweredZero",
			argLength: 3,
			reg: regInfo{
				inputs:   []regMask{buildReg("R16"), gp},
				clobbers: buildReg("R16"),
			},
			clobberFlags:   true,
			faultOnNilArg0: true,

	return uintptr(i1.x + i1.y + i2.x + i2.y + i3.x + i3.y + i4.x + i4.y + i5.x + i5.y)
}

// This test forces a GC. The idea is to have enough arguments
// that insufficient spill slots allocated (according to the ABI)
// may cause compiler-generated spills to clobber the return PC.
// Then, the GC stack scanning will catch that.
//
//go:registerparams
func sum9andGC(i1, i2, i3, i4, i5, i6, i7, i8, i9 uint32) uintptr {
	runtime.GC()

could discard <code>i</code> without saving a copy and then reload
<code>i = *p</code> just before
<code>funcs[i]()</code>.
A Go compiler must not, because the value of <code>*p</code> may have changed.
(Instead, the compiler could spill <code>i</code> to the stack.)
</p>

<p>
Not allowing a single write to write multiple values also means not using
the memory where a local variable will be written as temporary storage before the write.

goodm:
	MOVQ	R14, AX		// AX (and arg 0) = g
	MOVQ	SI, R14		// g = g.m.g0
	get_tls(CX)		// Set G in TLS
	MOVQ	R14, g(CX)
	MOVQ	(g_sched+gobuf_sp)(R14), SP	// sp = g0.sched.sp
	PUSHQ	AX	// open up space for fn's arg spill slot
	MOVQ	0(DX), R12
	CALL	R12		// fn(g)
	// The Windows native stack unwinder incorrectly classifies the next instruction
	// as part of the function epilogue, producing a wrong call stack.

	if framesize == 0 {
		return p
	}

	if ctxt.Flag_maymorestack != "" {
		// Save LR and REGCTXT
		const frameSize = 16
		p = ctxt.StartUnsafePoint(p, newprog)

		// Spill Arguments. This has to happen before we open
		// any more frame space.
		p = cursym.Func().SpillRegisterArgs(p, newprog)

		// MOV LR, -16(SP)
		p = obj.Appendp(p, newprog)
		p.As = AMOV

	abid = newAbiDesc(t, rcvr)

	// build dummy rtype holding gc program
	x := &abi.Type{
		Align_: goarch.PtrSize,
		// Don't add spill space here; it's only necessary in
		// reflectcall's frame, not in the allocated frame.
		// TODO(mknyszek): Remove this comment when register
		// spill space in the frame is no longer required.
		Size_:    align(abid.retOffset+abid.ret.stackBytes, goarch.PtrSize),

//     MOVL    DX, 4(DI)
// dowrite:
//     MOVL    AX, 88(CX)
TEXT gcWriteBarrier<>(SB),NOSPLIT,$28
	// Save the registers clobbered by the fast path. This is slightly
	// faster than having the caller spill these.
	MOVL	CX, 20(SP)
	MOVL	BX, 24(SP)
retry:
	// TODO: Consider passing g.m.p in as an argument so they can be shared
	// across a sequence of write barriers.
	get_tls(BX)
	MOVL	g(BX), BX

	// We don't save all registers on ppc64 because it takes too much space.
	MOVD	R20, (FIXED_FRAME+0)(R1)
	MOVD	R21, (FIXED_FRAME+8)(R1)
	// R0 is always 0, so no need to spill.
	// R1 is SP.
	// R2 is SB.
	MOVD	R3, (FIXED_FRAME+16)(R1)
	MOVD	R4, (FIXED_FRAME+24)(R1)
	MOVD	R5, (FIXED_FRAME+32)(R1)
	MOVD	R6, (FIXED_FRAME+40)(R1)
	MOVD	R7, (FIXED_FRAME+48)(R1)

		if !isLive(off, slotIdx) {
			print("?")
		}
	}

	start := true
	printcomma := func() {
		if !start {
			print(", ")
		}
	}
	pi := 0
	slotIdx := uint8(0) // register arg spill slot index
printloop:
	for {
		o := p[pi]
		pi++
		switch o {
		case abi.TraceArgsEndSeq:
			break printloop
		case abi.TraceArgsStartAgg:
			printcomma()
			print("{")
			start = true