b0.Event("ProcStart", trace.ProcID(0), testgen.Seq(1))
	b0.Event("GoStatus", trace.GoID(4), trace.NoThread, go122.GoRunnable)
	b0.Event("GoStart", trace.GoID(4), testgen.Seq(1))
	b0.Event("GoSyscallBegin", testgen.Seq(2), testgen.NoStack)
	b0.Event("GoDestroySyscall")

	// A new Go-created thread with the same ID appears and
	// starts running, then tries to steal the P from the
	// first thread. The stealing is interesting because if

	// on the system stack.
	SUB	$(asmcgocallSaveOffset+16), R1, R10
	RLDCR	$0, R10, $~15, R1	// 16-byte alignment for gcc ABI
	MOVD	R5, (asmcgocallSaveOffset+8)(R1)	// save old g on stack
	MOVD	(g_stack+stack_hi)(R5), R5
	SUB	R7, R5
	MOVD	R5, asmcgocallSaveOffset(R1)    // save depth in old g stack (can't just save SP, as stack might be copied during a callback)
#ifdef GOOS_aix

	MOVQ	m_g0(DI), DX		// g

	// Make TLS entries point at g and m.
	get_tls(BX)
	MOVQ	DX, g(BX)
	MOVQ	DI, g_m(DX)

	// Layout new m scheduler stack on os stack.
	MOVQ	SP, AX
	MOVQ	AX, (g_stack+stack_hi)(DX)
	SUBQ	$(0x100000), AX		// stack size
	MOVQ	AX, (g_stack+stack_lo)(DX)
	ADDQ	$const_stackGuard, AX
	MOVQ	AX, g_stackguard0(DX)
	MOVQ	AX, g_stackguard1(DX)

John Howard <******@****.***> 1710626582 -0700

  // TODO(apassos) multiple threads could be dequeuing from op_stack at the same
  // time, for better CPU backprop performance.
  VLOG(1) << "Initial stack:";
  if (VLOG_IS_ON(1)) {
    for (auto t : op_stack) {
      VLOG(1) << "  " << t;
    }
  }
  while (!op_stack.empty()) {
    const int64_t op = op_stack.back();
    VLOG(1) << "Popped " << op;
    op_stack.pop_back();
    auto op_it = state.op_tape.find(op);

	// Switch to system stack.
	// The original frame pointer is stored in BP,
	// which is useful for stack unwinding.
	CALL	gosave_systemstack_switch<>(SB)
	MOVQ	SI, g(CX)
	MOVQ	(g_sched+gobuf_sp)(SI), SP

	// Now on a scheduling stack (a pthread-created stack).
	SUBQ	$16, SP
	ANDQ	$~15, SP	// alignment for gcc ABI
	MOVQ	DI, 8(SP)	// save g
	MOVQ	(g_stack+stack_hi)(DI), DI
	SUBQ	DX, DI

    argument_set.insert(argument.getImpl());
  }

  // Searching for ops from results to arguments. Duplicate ops in the op stack
  // are intentional in order to make sure the op on the top of the stack
  // doesn't depends on any ops below it.
  std::stack<Operation*> op_stack;
  while (!value_queue.empty()) {
    Value current_value = value_queue.front();
    value_queue.pop();

	MOVD	m_g0(R0), g
	MOVD	R0, g_m(g)
	BL	runtime·save_g(SB)

	// Set up stack guards for OS stack.
	MOVD	RSP, R0
	MOVD	R0, g_stack+stack_hi(g)
	SUB	$(64*1024), R0
	MOVD	R0, (g_stack+stack_lo)(g)
	MOVD	R0, g_stackguard0(g)
	MOVD	R0, g_stackguard1(g)

	BL	runtime·emptyfunc(SB)	// fault if stack check is wrong
	BL	runtime·mstart(SB)

	RESTORE_R19_TO_R28(8*3)

	// Exit the thread.

	MOVW	g, m_g0(R8)
	// save g->m = m0
	MOVW	R8, g_m(g)

	// create istack out of the OS stack
	// (1MB of system stack is available on iOS and Android)
	MOVW	$(-64*1024+104)(R13), R0
	MOVW	R0, g_stackguard0(g)
	MOVW	R0, g_stackguard1(g)
	MOVW	R0, (g_stack+stack_lo)(g)
	MOVW	R13, (g_stack+stack_hi)(g)

	BL	runtime·emptyfunc(SB)	// fault if stack check is wrong

#ifdef GOOS_openbsd

	MOVW	m_g0(R0), g
	MOVW	R0, g_m(g)
	BL	runtime·save_g(SB)

	// Layout new m scheduler stack on os stack.
	MOVW	R13, R0
	MOVW	R0, g_stack+stack_hi(g)
	SUB	$(64*1024), R0
	MOVW	R0, (g_stack+stack_lo)(g)
	MOVW	R0, g_stackguard0(g)
	MOVW	R0, g_stackguard1(g)

	BL	runtime·emptyfunc(SB)	// fault if stack check is wrong
	BL	runtime·mstart(SB)

	// Exit the thread.
	MOVW	$0, R0