virtual absl::Status SetAttrInt(const char* attr_name, int64_t value) = 0;
  virtual absl::Status SetAttrFloat(const char* attr_name, float value) = 0;
  virtual absl::Status SetAttrBool(const char* attr_name, bool value) = 0;
  virtual absl::Status SetAttrType(const char* attr_name, DataType value) = 0;
  virtual absl::Status SetAttrShape(const char* attr_name, const int64_t* dims,
                                    const int num_dims) = 0;

    const TF_Operation* const* opers, int ninputs, const TF_Output* inputs,
    int noutputs, const TF_Output* outputs, const char* const* output_names,
    int ncontrol_outputs, const TF_Operation* const* control_outputs,
    const char* const* control_output_names, const TF_FunctionOptions* opts,
    const char* description, TF_Status* status) {
  tensorflow::mutex_lock l(fn_body->mu);

  // Process inputs.

    TFE_MonitoringStringGauge2* gauge, const char* label1, const char* label2) {
  return static_cast<TFE_MonitoringStringGaugeCell*>(
      static_cast<void*>(gauge->gauge->GetCell(label1, label2)));
}

TFE_MonitoringStringGauge3* TFE_MonitoringNewStringGauge3(
    const char* name, TF_Status* status, const char* description,
    const char* label1, const char* label2, const char* label3) {

				this.mem.setUint32(addr, id, true);
			}

			const loadSlice = (addr) => {
				const array = getInt64(addr + 0);
				const len = getInt64(addr + 8);
				return new Uint8Array(this._inst.exports.mem.buffer, array, len);
			}

			const loadSliceOfValues = (addr) => {
				const array = getInt64(addr + 0);
				const len = getInt64(addr + 8);
				const a = new Array(len);
				for (let i = 0; i < len; i++) {

absl::optional<std::vector<std::unique_ptr<ParallelTensor>>>
ParallelDevice::Execute(TFE_Context* context,
                        const std::vector<ParallelTensor*>& inputs,
                        const char* operation_name,
                        const TFE_OpAttrs* attributes, int expected_max_outputs,
                        TF_Status* status) const {
  std::vector<PartialTensorShape> expected_output_shapes(expected_max_outputs);

                                            TF_Status* status) const;

  // The number of devices operations run on.
  size_t num_underlying_devices() const { return underlying_devices_.size(); }

  // The devices operations run on.
  const std::vector<std::string>& underlying_devices() const {
    return underlying_devices_;
  }

  // Takes a description of a single operation being executed on the

}

void TF_AttrBuilderSetTypeList(TF_AttrBuilder* builder, const char* attr_name,
                               const TF_DataType* values, int num_values) {
  auto iter = builder->attr_names.insert(attr_name).first;
  builder->Set(*iter, tensorflow::gtl::ArraySlice<const tensorflow::DataType>(
                          reinterpret_cast<const tensorflow::DataType*>(values),
                          num_values));
}

  // that the argument need to be handled by a custom device.
  virtual bool HasCustomDeviceInput() const = 0;

  virtual const tensorflow::OpDef* OpDef() const = 0;

  virtual absl::Status InputLength(const char* input_name, int* length) = 0;
  virtual absl::Status OutputLength(const char* output_name, int* length) = 0;

  // Set stack trace to be used for potential async error reporting.

                       GradientFunction* gradient_function,
                       const string& op_name = "");
  // Returns whether any tensor in a list of tensors is being watched and has
  // a trainable dtype.
  bool ShouldRecord(
      absl::Span<const AbstractTensorHandle* const> tensors) const;
  // Unwatches this tensor on the tape. Mainly used for cleanup when deleting
  // eager tensors.

      const FunctionDef& fdef, const StackTracesMap& stack_traces) = 0;

  // Find and return a added function by its name.
  virtual const FunctionDef* FindFunctionDef(const string& name) const = 0;

  // Find and return a function record added by its name.
  virtual core::RefCountPtr<FunctionRecord> FindRecord(
      const string& name) const = 0;