#include "tensorflow/c/conversion_macros.h"
#include "tensorflow/c/eager/immediate_execution_operation.h"

// Wraps a pointer to an operation implementation.
//
// WARNING: Since the underlying object could be ref-counted a user of this
// interface cannot destruct the underlying operation object. Instead, call
// TFE_DeleteOp who calls Release() on the operation pointer and deletes
// the TFE_Op structure.
typedef struct TFE_Op TFE_Op;

#include "tensorflow/c/conversion_macros.h"
#include "tensorflow/c/eager/immediate_execution_tensor_handle.h"

// Wraps a pointer to a tensor handle implementation.
//
// WARNING: Since the underlying object could be ref-counted a user of this
// interface cannot destruct the underlying handle object. Instead, call
// TFE_DeleteTensorHandle who calls Release() on the handle pointer and deletes
// the TFE_TensorHandle structure.

// Flags indicating whether existing optimizers should be turned off.
// It's optional for plugin to set functions to return true/false. If not
// set, proper uses configuration set by user.
typedef struct TP_OptimizerConfigs {
  size_t struct_size;
  void* ext;  // reserved for future use
  TF_TriState disable_model_pruning;
  TF_TriState implementation_selector;
  TF_TriState function_optimization;
  TF_TriState common_subgraph_elimination;

      const = 0;
  virtual Status SetInput(size_t index,
                          ImmediateExecutionTensorHandle* input) = 0;

  virtual ImmediateExecutionContext* GetContext() const = 0;

  // Following two methods are used to support custom device.
  // Return true if the inputs contain custom device tensor handle. It means
  // that the argument need to be handled by a custom device.
  virtual bool HasCustomDeviceInput() const = 0;

    TFE_MonitoringSampler2* sampler, const char* label1, const char* label2);

// Sets whether to use TFRT
TF_CAPI_EXPORT extern void TFE_ContextOptionsSetTfrt(TFE_ContextOptions*,
                                                     bool use_tfrt);

// Returns the context_id from the EagerContext which is used by the
// EagerService to maintain consistency between client and worker. The

typedef struct TF_AbstractOp TF_AbstractOp;

// Stores a function representation that can be used for execution or for
// setting functional attributes of other composite ops e.g. control flow.
typedef struct TF_AbstractFunction TF_AbstractFunction;

// This allows the client to swap the implementation of the tracing engine.
// Any future call to TF_CreateFunction will use the implementation defined
// here.

#include "tensorflow/c/conversion_macros.h"
#include "tensorflow/c/eager/immediate_execution_context.h"

// Wraps a pointer to a context implementation.
//
// WARNING: Since the underlying object could be ref-counted a user of this
// interface cannot destruct the underlying context object. Instead, call
// TFE_DeleteContext who calls Release() on the context pointer and deletes
// the TFE_Context structure.
typedef struct TFE_Context TFE_Context;

  // session has been run (this is detected in RecordMutation function). If the
  // string is empty, no mutation has occurred. Otherwise the string is a
  // description of the mutation suitable for returning to the user.
  //
  // Sessions are added to this map in TF_NewSession, and removed in
  // TF_DeleteSession.
  // TF_Graph may only / must be deleted when
  //   sessions.size() == 0 && delete_requested == true
  //

        grad_it->second.push_back(grad);
      }
      result[i] = grad_it->second[0];
      used_gradient_ids.insert(tensor_id);
    }
  }
  VLOG(1) << "Final gradients size: "
          << gradients.size() - used_gradient_ids.size();
  for (const auto& grad_pair : gradients) {
    if (used_gradient_ids.find(grad_pair.first) == used_gradient_ids.end()) {
      for (const auto& g : grad_pair.second) {
        vspace.DeleteGradient(g);

typedef struct TF_Thread TF_Thread;

typedef struct TF_ThreadOptions {
  // Thread stack size to use (in bytes), zero implies that the system default
  // will be used.
  size_t stack_size;

  // Guard area size to use near thread stacks to use (in bytes), zero implies
  // that the system default will be used.
  size_t guard_size;

  // The NUMA node to use, -1 implies that there should be no NUMA affinity for
  // this thread.
  int numa_node;