// Create a variable handle (uninitialized to start) placed on the parallel
  // device.
  std::function<void(Variable*)> variable_deleter = [&](Variable* to_delete) {
    to_delete->Destroy(context, status.get());
    ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
    delete to_delete;
  };
  std::unique_ptr<Variable, decltype(variable_deleter)> variable(

namespace tensorflow {
namespace parallel_device {

// A helper for performing common operations on variables. A much more
// restricted stand-in for tf.Variable in Python.
class Variable {
 public:
  // Construct a Variable from a resource-dtype TFE_TensorHandle and an
  // indication of the dtype of the variable's value.
  //
  // Note that creating this resource-dtype handle can fail, so `Create` is a

# sets the __all__ variable. If it does, we have to be sure to add
# "contrib".
if "__all__" in vars():
  vars()["__all__"].append("contrib")

from tensorflow.python.platform import flags
# The "app" module will be imported as part of the placeholder section above.
_current_module.app.flags = flags  # pylint: disable=undefined-variable
setattr(_current_module, "flags", flags)

       * Refer [windows setup guide](https://www.tensorflow.org/install/gpu#windows_setup).
     * If you have above configuration and using _**Ubuntu/Linux**_ platform -
       * Try adding the CUDA, CUPTI, and cuDNN installation directories to the $LD_LIBRARY_PATH environment variable.
       * Refer [linux setup guide](https://www.tensorflow.org/install/gpu#linux_setup).

    TF_CheckpointReader* reader, const char* name);
// Get the variable name at the given index
TF_CAPI_EXPORT extern const char* TF_CheckpointReaderGetVariable(
    TF_CheckpointReader* reader, int index);
// Get the number of variable in the checkpoint
TF_CAPI_EXPORT extern int TF_CheckpointReaderSize(TF_CheckpointReader* reader);
// Get the DataType of a variable
TF_CAPI_EXPORT extern TF_DataType TF_CheckpointReaderGetVariableDataType(

TFE_TensorHandle* CreateVarHandle(TFE_Context* ctx,
                                  const tensorflow::string& device_name,
                                  const tensorflow::string& variable_name) {
  TF_Status* status = TF_NewStatus();
  // Create the variable handle.
  TFE_Op* op = TFE_NewOp(ctx, "VarHandleOp", status);
  if (TF_GetCode(status) != TF_OK) return nullptr;
  TFE_OpSetAttrType(op, "dtype", TF_FLOAT);

  /// the LRU cache, the timestamp (seconds since epoch) at which the block
  /// was cached, a coordination lock, and state & condition variables.
  ///
  /// Thread safety:
  /// The iterator and timestamp fields should only be accessed while holding
  /// the block-cache-wide mu_ instance variable. The state variable should only
  /// be accessed while holding the Block's mu lock. The data vector should only

// Create a variable handle with name `variable_name` on a device with name
// `device_name`.
TFE_TensorHandle* CreateVarHandle(TFE_Context* ctx,
                                  const tensorflow::string& device_name,
                                  const tensorflow::string& variable_name);

// Create a variable with value `value` and name `variable_name` on a device
// with name `device_name`.

# variables in it, OR may be left empty if the user has already exported the
# relevant variables in their environment. Because of 'set -o allexport' above
# (which is equivalent to "set -a"), every variable in the file is exported
# for other files to use.
#
# Separately, if TFCI is set *and* there are also additional TFCI_ variables

  tensorflow::Var* variable;
  cc_status = tensorflow::LookupResource(cc_ctx, handle, &variable);
  return new TF_VariableInfo(index, handle.name(), variable);
}

void TF_LockVariableInfos(TF_VariableInfo** vars, int num_vars,
                          TF_Status* status) {
  std::vector<tensorflow::VariableInfo*> variable_ptrs;
  variable_ptrs.reserve(num_vars);
  for (int i = 0; i < num_vars; ++i) {