Block& block = parallel_execute.GetRegionBlockWithIndex(i);
    auto results = parallel_execute.GetRegionOutputs(i);
    auto& results_to_remove = results_to_remove_by_region[i];
    results_to_remove.reserve(results.size());
    Operation* terminator = block.getTerminator();
    for (auto result : llvm::enumerate(results)) {
      TF::AssignVariableOp assign_var =
          GetSingleUseResourceWrite(parallel_execute, result.value());

	CBZ	R4, nocgo
	MOVD	$_rt0_arm64_netbsd_lib_go(SB), R0
	MOVD	$0, R1
	SUB	$16, RSP		// reserve 16 bytes for sp-8 where fp may be saved.
	BL	(R4)
	ADD	$16, RSP
	B	restore

nocgo:
	MOVD	$0x800000, R0                     // stacksize = 8192KB
	MOVD	$_rt0_arm64_netbsd_lib_go(SB), R1
	MOVD	R0, 8(RSP)
	MOVD	R1, 16(RSP)
	MOVD	$runtime·newosproc0(SB),R4
	BL	(R4)

restore:

  computeInputsOutputs(ops, &inputs, &outputs);

  std::vector<Type> input_types;
  std::vector<Type> output_types;

  input_types.reserve(inputs.size());
  for (Value v : inputs) {
    input_types.push_back(v.getType());
  }
  output_types.reserve(outputs.size());
  for (Value v : outputs) {
    output_types.push_back(v.getType());
  }

  // Create the function.

                        llvm::SmallVectorImpl<int64_t>* shape) {
  shape->reserve(input_shape.dims());
  for (const auto& d : input_shape) {
    shape->push_back(d.size == kTFDynamicSize ? ShapedType::kDynamic : d.size);
  }
}

Status ConvertToMlirShape(const TensorShapeProto& input_shape,
                          llvm::SmallVectorImpl<int64_t>* shape) {
  shape->reserve(input_shape.dim_size());
  auto& dims = input_shape.dim();

	CBZ	R4, nocgo
	MOVD	$_rt0_arm64_linux_lib_go(SB), R0
	MOVD	$0, R1
	SUB	$16, RSP		// reserve 16 bytes for sp-8 where fp may be saved.
	BL	(R4)
	ADD	$16, RSP
	B	restore

nocgo:
	MOVD	$0x800000, R0                     // stacksize = 8192KB
	MOVD	$_rt0_arm64_linux_lib_go(SB), R1
	MOVD	R0, 8(RSP)
	MOVD	R1, 16(RSP)
	MOVD	$runtime·newosproc0(SB),R4
	BL	(R4)

restore:
	// Restore callee-save registers.

	BL	(R4)

	// Create a new thread to do the runtime initialization and return.
	MOVD	_cgo_sys_thread_create(SB), R4
	CBZ	R4, nocgo
	MOVD	$_rt0_arm64_openbsd_lib_go(SB), R0
	MOVD	$0, R1
	SUB	$16, RSP		// reserve 16 bytes for sp-8 where fp may be saved.
	BL	(R4)
	ADD	$16, RSP
	B	restore

nocgo:
	MOVD	$0x800000, R0                     // stacksize = 8192KB
	MOVD	$_rt0_arm64_openbsd_lib_go(SB), R1
	MOVD	R0, 8(RSP)

	MOVD	_cgo_sys_thread_create(SB), R4
	CBZ	R4, nocgo
	MOVD	$_rt0_arm64_freebsd_lib_go(SB), R0
	MOVD	$0, R1
	SUB	$16, RSP	// reserve 16 bytes for sp-8 where fp may be saved.
	BL	(R4)
	ADD	$16, RSP
	B	restore

nocgo:
	MOVD	$0x800000, R0                     // stacksize = 8192KB
	MOVD	$_rt0_arm64_freebsd_lib_go(SB), R1
	MOVD	R0, 8(RSP)
	MOVD	R1, 16(RSP)
	MOVD	$runtime·newosproc0(SB),R4

                                  const std::set<int>* variables_updated,
                                  std::vector<VariableInfo>* result) {
  result->clear();
  result->reserve(variable_indices.size());
  for (int var_idx : variable_indices) {
    Var* variable = nullptr;
    if (inputs[var_idx]->NumElements() == 0) {
      return errors::InvalidArgument("Empty resource tensor passed  at index ",

  if (!session) return module.emitOpError() << "no session provided";

  // Read all resource variables from the session.
  std::vector<std::string> variable_names;
  variable_names.reserve(resource_names.size());
  for (StringRef name : resource_names) variable_names.push_back(name.str());

  std::vector<Tensor> resource_tensors;
  Status status = session->Run(
      /*inputs=*/{}, variable_names,

    return (emitError(loc, "unsupported FakeQuant number of bits: ") << numBits,
            nullptr);
  }

  SmallVector<double, 4> scales;
  SmallVector<int64_t, 4> zeroPoints;
  scales.reserve(axisSize);
  zeroPoints.reserve(axisSize);
  for (size_t axis = 0; axis != axisSize; ++axis) {
    double rmin = rmins[axis];
    double rmax = rmaxs[axis];
    if (std::fabs(rmax - rmin) < std::numeric_limits<double>::epsilon()) {