"important to you");
  return EmptyWhileParams();
#else
  if (ninputs == 0) {
    status->status =
        InvalidArgument("TF_NewWhile() must be passed at least one input");
    return EmptyWhileParams();
  }

  TF_Graph* cond_graph = TF_NewGraph();
  TF_Graph* body_graph = TF_NewGraph();
  cond_graph->parent = g;
  cond_graph->parent_inputs = inputs;

  // a tensor with the result.
  AbstractTensorHandle* AggregateGradients(
      gtl::ArraySlice<AbstractTensorHandle*> gradient_tensors) const override;

  // Calls the passed-in backward function.
  // op_type is the op's name provided in RecordOperation.
  Status CallBackwardFunction(
      const string& op_type, GradientFunction* gradient_function,

}

// For TFE_CustomDevice::copy_tensor_to_device in the parallel device
// registration.
//
// Since this function is used to satisfy the TFE_CustomDevice C API,
// device_info is passed in using a C-style generic. It must always be a
// ParallelDevice.
TFE_TensorHandle* CopyToParallelDevice(TFE_Context* context,
                                       TFE_TensorHandle* tensor,

        }));
  }
  // The `threads` destructor blocks until the threads can be joined, once their
  // respective reads finish (which happens once they are all concurrently being
  // executed, or 10 seconds have passed).
}

TEST(RamFileBlockCacheTest, CoalesceConcurrentReads) {
  // Concurrent reads to the same file blocks should be de-duplicated.
  const size_t block_size = 16;
  int num_requests = 0;

}

void Variable::Assign(TFE_Context* context, TFE_TensorHandle* value,
                      TF_Status* status) {
  GeneralAssignment("AssignVariableOp", context, value, status);
}

// Passed to `TF_NewTensor` to indicate how an array of floats should be
// deleted.
static void FloatDeallocator(void* data, size_t, void* arg) {
  delete[] static_cast<float*>(data);
}

}

// -------------------------  VariableInfo  ------------------------------------
struct TF_VariableInfo {
  TF_VariableInfo() = delete;
  // TF_VariableInfo is constructed here by TensorFlow, and will be passed to
  // plugin as a opaque pointer. Plugin will need to call C APIs below to
  // operate on TF_VariableInfo (such as allocate temp tensor for the `var` held
  // by the underlying tensorflow::VariableInfo.

                                  TF_Status* status) {
  tensorflow::AttrValue attr_value;
  if (!attr_value.ParseFromArray(proto, proto_len)) {
    status->status = InvalidArgument(
        "Unparseable AttrValue proto passed to "
        "TF_FunctionSetAttrValueProto");
    return;
  }

  auto fdef_or = func->record->mutable_fdef();
  if (!fdef_or.ok()) {
    status->status = fdef_or.status();
    return;
  }

// become available after a shared object with a filesystem implementation is
// loaded, we can only check for availability after all arguments have been
// parsed.
//
// Furthermore, as `INSTANTIATE_TEST_SUITE_P` needs to be at global level and we
// don't want to have a `std::vector<std::string>` at global level, we use a

  const char* device_name =
      tensorflow::unwrap(h)->BackingDeviceName(&status->status);
  DeviceNameUtils::ParsedName parsed_name;
  tensorflow::DeviceNameUtils::ParseFullName(device_name, &parsed_name);
  std::string device_type = parsed_name.type;
  int device_id = 0;
  if (parsed_name.has_id) {
    device_id = parsed_name.id;
  }

  ctx.device_id = device_id;
  if (device_type == "CPU") {

    }
  }
  std::vector<std::string> local_names;
  local_names.reserve(underlying_devices_.size());
  for (const DeviceNameUtils::ParsedName& parsed_component :
       parsed_components) {
    local_names.push_back(
        absl::StrCat(parsed_component.type, ":", parsed_component.id));
  }
  return local_names;
}

std::unique_ptr<ParallelTensor> ParallelTensor::FromTensorHandles(