// Enables only graph collection in RunMetadata on the functions executed from
// this context.
TF_CAPI_EXPORT extern void TFE_ContextEnableGraphCollection(TFE_Context* ctx);

// Disables only graph collection in RunMetadata on the functions executed from
// this context.
TF_CAPI_EXPORT extern void TFE_ContextDisableGraphCollection(TFE_Context* ctx);

// TODO(fishx): Move these monitoring APIs into a separate file.

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

// Return a matmul op multiplying `a` by `b`.
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);

  template <typename DataType>
  std::unique_ptr<ParallelTensor> ScalarsFromSequence(
      absl::Span<const DataType> values, TFE_Context* context,
      TF_Status* status) const;

  // A parallel tensor with scalar integers numbering component devices.
  std::unique_ptr<ParallelTensor> DeviceIDs(TFE_Context* context,
                                            TF_Status* status) const;

  // The number of devices operations run on.

// instead of the original node's output.
//
// Callers must also satisfy the following constraints:
// - `inputs` cannot refer to TF_Outputs within a control flow context. For
//   example, one cannot use the output of "switch" node as input.
// - `inputs` and `outputs` cannot have reference types. Reference types are

TF_CAPI_EXPORT extern void TFE_AbortCollectiveOps(TFE_Context* ctx,
                                                  TF_Status* status);

// Checks the health of collective ops peers. Explicit health check is needed in
// multi worker collective ops to detect failures in the cluster.  If a peer is
// down, collective ops may hang.
TF_CAPI_EXPORT extern void TFE_CollectiveOpsCheckPeerHealth(
    TFE_Context* ctx, const char* task, int64_t timeout_in_ms,

  bool IsPersistent() const { return persistent_; }

 private:
  TensorTape tensor_tape_;
  OpTape<BackwardFunction, TapeTensor> op_tape_;
  int64_t next_op_id_{0};

  // Map from tensor id to number of remaining usages (i.e. how many entries in
  // the tape refer to it); to aid in tape garbage collection.
  std::unordered_map<int64_t, int64_t> tensor_usage_;

//
// TODO(ashankar): Merge with TF_Session?
typedef struct TFE_Context TFE_Context;

TF_CAPI_EXPORT extern TFE_Context* TFE_NewContext(
    const TFE_ContextOptions* opts, TF_Status* status);
TF_CAPI_EXPORT extern void TFE_DeleteContext(TFE_Context* ctx);
TF_CAPI_EXPORT extern TF_DeviceList* TFE_ContextListDevices(TFE_Context* ctx,
                                                            TF_Status* status);

  virtual ~ImmediateExecutionDistributedManager() {}

  // Set up distributed execution environment on local and remote tasks.
  // When `reset_context` is true, initialize new cluster context state based
  // on cluster configurations provided in `server_def`; otherwise, update
  // existing context state with the provided `server_def`. Contexts created
  // on remote tasks will be considered stale and garbage collected after

#ifndef TENSORFLOW_C_EAGER_IMMEDIATE_EXECUTION_CONTEXT_H_
#define TENSORFLOW_C_EAGER_IMMEDIATE_EXECUTION_CONTEXT_H_

#include <functional>
#include <map>
#include <memory>
#include <vector>

#include "absl/types/optional.h"
#include "absl/types/span.h"
#include "tensorflow/c/eager/abstract_context.h"
#include "tensorflow/c/eager/immediate_execution_distributed_manager.h"

#ifndef TENSORFLOW_C_EAGER_GRADIENTS_H_
#define TENSORFLOW_C_EAGER_GRADIENTS_H_

#include "absl/container/flat_hash_map.h"
#include "tensorflow/c/eager/abstract_context.h"
#include "tensorflow/c/eager/abstract_tensor_handle.h"
#include "tensorflow/c/eager/tape.h"
#include "tensorflow/core/common_runtime/eager/attr_builder.h"

namespace tensorflow {
namespace gradients {