#endif  // !defined(TENSORFLOW_NO_SHARED_OBJECTS)
#endif  // !defined(PLATFORM_WINDOWS)
}

void DefineFunction(const char* name, TF_Function** func,
                    const char* description = nullptr,
                    bool append_hash = false) {
  std::unique_ptr<TF_Graph, decltype(&TF_DeleteGraph)> func_graph(
      TF_NewGraph(), TF_DeleteGraph);
  std::unique_ptr<TF_Status, decltype(&TF_DeleteStatus)> s(TF_NewStatus(),

    `SparseTensor` arguments.
*   The Python bitwise operators for `Tensor` (`__and__`, `__or__`, `__xor__`
    and `__invert__` now support non-`bool` arguments and apply the
    corresponding bitwise ops. `bool` arguments continue to be supported and
    dispatch to logical ops. This brings them more in line with Python and NumPy
    behavior.

  // true if async execution is enabled.
  bool async = false;
  TFE_ContextDevicePlacementPolicy device_placement_policy{
      TFE_DEVICE_PLACEMENT_SILENT};
  // If true, use TFRT backend
  bool use_tfrt = false;
  // Whether to run elementary eager ops wrapped in a call op.
  bool run_eager_op_as_function = false;
  // Whether to rewrite jit_compile functions.
  bool jit_compile_rewrite = false;
};

  return ops::DivNoNan(ctx, inputs[0], inputs[1], &outputs[0], "DivNoNan");
}

class CppGradients
    : public ::testing::TestWithParam<std::tuple<const char*, bool, bool>> {
 protected:
  void SetUp() override {
    TF_StatusPtr status(TF_NewStatus());
    TF_SetTracingImplementation(std::get<0>(GetParam()), status.get());
    status_ = StatusFromTF_Status(status.get());

  explicit GradientTape(bool persistent) : persistent_(persistent) {}
  ~GradientTape() {
    for (const auto& pair : op_tape_) {
      pair.second.backward_function_deleter(pair.second.backward_function);
    }
  }

  // Returns whether any tensor in a list of tensors is being watched and has
  // a trainable dtype.
  bool ShouldRecord(absl::Span<const int64_t> tensor_ids,

namespace tensorflow {
namespace gradients {
namespace internal {
namespace {
using std::vector;

class CustomGradientTest
    : public ::testing::TestWithParam<std::tuple<const char*, bool, bool>> {
 protected:
  void SetUp() override {
    TF_StatusPtr status(TF_NewStatus());
    TF_SetTracingImplementation(std::get<0>(GetParam()), status.get());
    Status s = StatusFromTF_Status(status.get());

  TFE_Execute(op, out_value, &num_retvals, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_DeleteOp(op);
  TF_DeleteStatus(status);
}

void TestRemoteExecuteChangeServerDef(bool async) {
  tensorflow::ServerDef server_def = GetServerDef(2);

  // This server def has the task index set to 0.
  string serialized = server_def.SerializeAsString();

  server_def.set_task_index(1);

      tensorflow::GetBuildXlaOpsPassFlags();
  bool original = flags->tf_xla_enable_lazy_compilation;
  flags->tf_xla_enable_lazy_compilation = enable;
  return original;
}

unsigned char TF_SetTfXlaCpuGlobalJit(unsigned char enable) {
  tensorflow::MarkForCompilationPassFlags* flags =
      tensorflow::GetMarkForCompilationPassFlags();
  bool original = flags->tf_xla_cpu_global_jit;

namespace tensorflow {
namespace gradients {
namespace internal {

void CompareNumericalAndAutodiffGradients(
    Model model, Model grad_model, AbstractContext* ctx,
    absl::Span<AbstractTensorHandle* const> inputs, bool use_function,
    double abs_error) {
  auto num_inputs = inputs.size();
  std::vector<AbstractTensorHandle*> outputs(num_inputs);
  auto s = RunModel(grad_model, ctx, inputs, absl::MakeSpan(outputs),

    error_msg: (String) String with one and only one '%s'. Formatted with each
      invalid response upon check_success(input) failure.
    suppress_default_error: (Bool) Suppress the above error message in favor of
      one from the check_success function.
    resolve_symlinks: (Bool) Translate symbolic links into the real filepath.
    n_ask_attempts: (Integer) Number of times to query for valid input before
      raising an error and quitting.