// devices of a ParallelDevice. If called, ParallelTensor::Shape inspects
  // `components` to determine a shape.
  static std::unique_ptr<ParallelTensor> FromTensorHandles(
      const ParallelDevice& parallel_device,
      std::vector<TensorHandlePtr> components, TF_Status* status);
  // Uses the provided shape without additional checks, which avoids blocking
  // when ParallelTensor::Shape is called.

// setting a shape of [-1, 2] with an existing shape [2, -1] would set
// a final shape of [2, 2] based on shape merging semantics.
//
// Returns an error into `status` if:
//   * `output` is not in `graph`.
//   * An invalid shape is being set (e.g., the shape being set
//     is incompatible with the existing shape).
TF_CAPI_EXPORT extern void TF_GraphSetTensorShape(TF_Graph* graph,

TFE_Op* MatMulOp(TFE_Context* ctx, TFE_TensorHandle* a, TFE_TensorHandle* b);

// Return an identity op.
TFE_Op* IdentityOp(TFE_Context* ctx, TFE_TensorHandle* a);

// Return a shape op fetching the shape of `a`.
TFE_Op* ShapeOp(TFE_Context* ctx, TFE_TensorHandle* a);

// Return an allreduce op adding up input tensor `in` from `group_size` workers.

// per device in `underlying_devices`. Implicit copies off of the device throw
// an error.
//
// All component tensors must have the same dtype. Currently they must also have
// the same shape, although this requirement may be relaxed in the future.
//
// `device_name` must not name an existing physical or custom device (see
// the documentation for TFE_RegisterCustomDevice for more information).
//

// fastbuild.
inline Status AbstractOperation::SetAttrShape(const char* attr_name,
                                              const PartialTensorShape shape) {
  return SetAttrShape(attr_name, shape.dim_sizes().data(), shape.dims());
}

inline Status AbstractOperation::SetAttrStringList(
    const char* attr_name, absl::Span<string const> values) {
  std::vector<const char*> raw_strs;

                                                   TF_Status* status);

// A handle to a tensor on a device.
//
// Like a TF_Tensor, a TFE_TensorHandle refers to a tensor with a value, shape,
// type etc. Unlike a TF_Tensor, a TFE_TensorHandle may refer to such tensors
// placed in the memory of different devices or remote address spaces.
typedef struct TFE_TensorHandle TFE_TensorHandle;

// Represents a (partially-defined) shape.
typedef struct TF_Shape {
  int num_dims;  // Must be >= -1; -1 represents unknown rank.
  int64_t* dim_sizes;
} TF_Shape;

// Add a new parameter to a TensorFlow Function.
TF_AbstractTensor* TF_AddFunctionParameter(TF_ExecutionContext* func,
                                           TF_DataType dtype, TF_Shape shape,
                                           TF_Status* s);

#include <stack>
#include <unordered_map>
#include <unordered_set>
#include <vector>

#include "tensorflow/core/config/flag_defs.h"
#include "tensorflow/core/config/flags.h"
#include "tensorflow/core/framework/tensor_shape.h"
#include "tensorflow/core/framework/types.h"
#include "tensorflow/core/lib/gtl/array_slice.h"
#include "tensorflow/core/lib/gtl/cleanup.h"
#include "tensorflow/core/lib/gtl/flatmap.h"

  TF_TriState function_optimization;
  TF_TriState common_subgraph_elimination;
  TF_TriState arithmetic_optimization;
  TF_TriState debug_stripper;
  TF_TriState constant_folding;
  TF_TriState shape_optimization;
  TF_TriState auto_mixed_precision;
  TF_TriState auto_mixed_precision_onednn_bfloat16;
  TF_TriState auto_mixed_precision_mkl;
  TF_TriState pin_to_host_optimization;
  TF_TriState layout_optimizer;

  // Computes the rank of the tensor handle.
  //
  // Shapes are specified via callbacks because retrieving the shape of a tensor
  // is a blocking operation for async eager; custom devices should avoid
  // retrieving shapes of tensors they wrap until the custom device tensor's
  // shape is explicitly requested where possible.
  int (*num_dims)(void* data, TF_Status* status);

  // Computes the axis length at `dim_index`.