// Returns tensor dtype.
  virtual DataType Type() const = 0;
  // Returns number of dimensions.
  virtual int NumDims() const = 0;
  // Returns size of specified dimension
  virtual int64_t Dim(int dim_index) const = 0;
  // Returns number of elements across all dimensions.
  virtual int64_t NumElements() const = 0;
  // Return size in bytes of the Tensor
  virtual size_t ByteSize() const = 0;

TF_CAPI_EXPORT extern int TFE_TensorDebugInfoOnDeviceNumDims(
    TFE_TensorDebugInfo* debug_info);

// Returns the number of elements in dimension `dim_index`.
// Tensor representation on device can be transposed from its representation
// on host. The data contained in dimension `dim_index` on device
// can correspond to the data contained in another dimension in on-host
// representation. The dimensions are indexed using the standard TensorFlow

  // ImmediateExecutionTensorHandles, but comes up for example when computing
  // the shape of a parallel tensor with component shapes differing across
  // devices.
  virtual Status Dim(int dim_index, int64_t* dim) const = 0;

  // Returns the device which created the handle.
  virtual const char* DeviceName(Status* status) const = 0;
  // Returns the device where the tensor was placed.

  ~TensorInterface() override {}

  void Release() override;

  DataType Type() const override;
  int NumDims() const override;
  int64_t Dim(int dim_index) const override;
  int64_t NumElements() const override;
  size_t ByteSize() const override;
  void* Data() const override;
  bool IsAligned() const override;
  bool CanMove() const override;

	// the 'begin' method at the start, then the 'end' method when
	// we're done with the file.
	BeginCounterDataFile(cdf string, cdr *decodecounter.CounterDataReader, dirIdx int)
	EndCounterDataFile(cdf string, cdr *decodecounter.CounterDataReader, dirIdx int)

	// Invoked once for each live function in the counter data file.
	VisitFuncCounterData(payload decodecounter.FuncPayload)

	WriteAll(ctx context.Context, volume string, path string, b []byte) (err error)

	// Read all.
	ReadAll(ctx context.Context, volume string, path string) (buf []byte, err error)
	GetDiskLoc() (poolIdx, setIdx, diskIdx int) // Retrieve location indexes.

      CompileTimeConstantOperand<TF::XlaReduceWindowOp, 2, 3, 4, 5, 6>,
      // $dim_index
      CompileTimeConstantOperand<TF::XlaRemoveDynamicDimensionSizeOp, 1>,
      // $window_dimensions, $window_strides, $padding
      CompileTimeConstantOperand<TF::XlaSelectAndScatterOp, 1, 2, 3>,
      CompileTimeConstantOperand<TF::XlaSetBoundOp, 1>,  // $bound
      // $dim_index
      CompileTimeConstantOperand<TF::XlaSetDynamicDimensionSizeOp, 1>

  return num_elements;
}

int64_t TFE_TensorHandleDim(TFE_TensorHandle* h, int dim_index,
                            TF_Status* status) {
  if (h == nullptr) {
    status->status = tensorflow::errors::InvalidArgument("Invalid handle");
    return -1;
  }

  int64_t dim = -1;
  status->status = tensorflow::unwrap(h)->Dim(dim_index, &dim);
  return dim;
}

// dimension of a parallel tensor.
int64_t ParallelTensorDim(void* data, int dim_index, TF_Status* status) {
  const std::vector<int64_t>* shape;
  Status s = reinterpret_cast<ParallelTensor*>(data)->Shape(&shape);
  if (!s.ok()) {
    tsl::Set_TF_Status_from_Status(status, s);
    return -1;
  }
  return (*shape)[dim_index];
}

TF_Buffer* ParallelTensorSummarize(void* data, TF_Status* status) {

                        const std::vector<Output>& grad_inputs,
                        std::vector<Output>* grad_outputs,
                        int start_value_index, int end_value_index,
                        int dim_index) {
  if (end_value_index >= op.num_inputs()) {
    return errors::Internal("Invalid input index");
  }
  std::vector<Output> inputs;
  inputs.reserve(end_value_index - start_value_index);