package main

type Element interface {
}

type Vector struct {
	elem []Element;
}

func (v *Vector) At(i int) Element {
	return v.elem[i];
}

type TStruct struct {
	name string;
	fields *Vector;
}

func (s *TStruct) field(i int) *TStruct {
	return s.fields.At(i).(*TStruct);
}

func main() {
	v := new(Vector);
	v.elem = make([]Element, 10);
	t := new(TStruct);
	t.name = "hi";

}

// Returns nodes with given type.
std::vector<Node*> GatherNodesWithType(const Graph& g, const string& type) {
  std::vector<Node*> result;
  for (Node* n : g.nodes()) {
    if (n->type_string() == type) {
      result.push_back(n);
    }
  }
  return result;
}

// Gets data types from `arg_nodes` and fills them into `recv_at_host_dtypes`.
Status GetArgDataTypes(const std::vector<Node*>& arg_nodes,

}

// -----

func.func @constant_tensor_invalid_1(%arg0: vector<1xi32>) -> tensor<?xi32> {
    // expected-error@+1 {{output type should be static and ranked}}
  %0 = "tfr.constant_tensor"(%arg0) : (vector<1xi32>) -> tensor<?xi32>
  func.return %0 : tensor<?xi32>
}

// -----

func.func @constant_tensor_invalid_2(%arg0: vector<1xi32>) -> tensor<1xf32> {

PruneChunkedMessage(
    const ::tensorflow::proto_splitter::ChunkedMessage& chunked_message,
    riegeli::RecordReader<riegeli::FdReader<>>& reader,
    std::vector<::tensorflow::proto_splitter::ChunkInfo> chunks_info,
    std::vector<RepeatedPtrField<::tensorflow::proto_splitter::FieldIndex>>
        target_fields_list);

// Deterministically serializes the proto `message`.

      /* cluster_flr= */ nullptr));
}

std::vector<DataType> DataTypeSetToVector(DataTypeSet set) {
  std::vector<DataType> result;
  result.reserve(set.size());
  for (DataType dt : set) {
    result.push_back(dt);
  }
  return result;
}

std::vector<std::vector<int64_t>> InterestingShapes() {
  std::vector<std::vector<int64_t>> interesting_shapes;

    llvm::StringRef device_type,
    XlaShapeLayoutHelpers::ShapeDeterminationFns shape_determination_fns,
    const std::vector<tensorflow::TensorShape>& arg_shapes,
    std::vector<tpu::ShardingAndIndex>* arg_core_mapping,
    std::vector<std::vector<xla::Shape>>* per_core_arg_shapes,
    std::vector<std::unique_ptr<mlir::Pass>>& custom_legalization_passes,
    xla::CompileOnlyClient* client, XlaCompilationResult* compilation_result) {

  // the gradient w.r.t. to 'x' (returned in grad_outputs[1]).
  auto dz = Const(scope_test_, {{5}});
  std::vector<Output> grad_outputs;
  TF_ASSERT_OK(
      AddSymbolicGradients(scope_test_, {z}, {v, x}, {dz}, &grad_outputs));

  std::vector<Tensor> outputs;
  test::GetTensors(scope_test_, {grad_outputs[0], grad_outputs[1]}, &outputs);

    thread_pool.Schedule([&session, a]() {
      std::vector<Tensor> outputs;
      TF_EXPECT_OK(session.Run({a}, &outputs));
      test::ExpectTensorEqual<int>(outputs[0],
                                   test::AsTensor<int>({4, 6}, {2}));
    });
    thread_pool.Schedule([&session, b]() {
      std::vector<Tensor> outputs;
      TF_EXPECT_OK(session.Run({b}, &outputs));

	MOVBZ (R_TMP), R_TMP       // load bool indicating support for vector facility
	CGIJ  $EQ, R_TMP, $0, novx // jump to scalar code if the vector facility is not available

	// Start of vector code (have vector facility).
	//
	// Set R_LEN to be the length mod 16 minus 1 to use as an index for
	// vector 'load with length' (VLL). It will be in the range [-1,14].
	// Also replicate c across a 16-byte vector and initialize V_ZERO.
	ANDW  $0xf, R_LEN

    llvm::StringRef device_type,
    std::vector<std::unique_ptr<mlir::Pass>>& custom_legalization_passes,
    XlaShapeLayoutHelpers::ShapeDeterminationFns shape_determination_fns,
    const std::vector<tensorflow::TensorShape>& arg_shapes,
    std::vector<tpu::ShardingAndIndex>* arg_core_mapping,
    std::vector<std::vector<xla::Shape>>* per_core_arg_shapes,
    xla::CompileOnlyClient* client);