const char* first_device = "/job:worker/replica:0/task:1/device:CPU:0";
  const char* second_device = "/job:worker/replica:0/task:2/device:CPU:0";
  const char* device_name = "/job:localhost/replica:0/task:0/device:CUSTOM:0";
  std::array<const char*, 2> underlying_devices{first_device, second_device};
  RegisterParallelDevice(context.get(), device_name, underlying_devices,
                         status.get());

  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  const DLTensor* dltensor_out = &dlm_out->dl_tensor;
  EXPECT_EQ(dltensor_out->device.device_type, dltensor_in->device.device_type);
  EXPECT_EQ(dltensor_out->device.device_id, dltensor_in->device.device_id);
  EXPECT_EQ(dltensor_out->ndim, dltensor_in->ndim);
  EXPECT_EQ(dltensor_out->dtype.code, dltensor_in->dtype.code);

}

void TF_SetRequestedDevice(TF_Graph* graph, TF_Operation* op,
                           const char* device) {
  using tensorflow::RecordMutation;
  mutex_lock l(graph->mu);
  op->node.set_requested_device(device);
  RecordMutation(graph, *op, "setting device");
}

void TF_RemoveAllControlInputs(TF_Graph* graph, TF_Operation* op) {
  mutex_lock l(graph->mu);

void TF_AbstractOpSetOpType(TF_AbstractOp* op, const char* const op_type,
                            TF_Status* s) {
  tsl::Set_TF_Status_from_Status(
      s, unwrap(op)->Reset(op_type,
                           /*raw_device_name=*/nullptr));
}

void TF_AbstractOpSetOpName(TF_AbstractOp* op, const char* const op_name,
                            TF_Status* s) {
  TracingOperation* tracing_op = dyn_cast<TracingOperation>(unwrap(op));

#include "tensorflow/core/tfrt/common/pjrt_util.h"
#include "tsl/distributed_runtime/coordination/coordination_service_agent.h"

TF_Device* TF_GetDevice(TF_OpKernelContext* ctx) {
  auto* cc_ctx = reinterpret_cast<tensorflow::OpKernelContext*>(ctx);
  return reinterpret_cast<TF_Device*>(cc_ctx->device());
}

// --------------------------  Resource  ---------------------------------------

#include "tensorflow/c/eager/parallel_device/parallel_device_lib.h"

#include "tensorflow/c/c_api.h"
#include "tensorflow/c/c_api_experimental.h"
#include "tensorflow/c/eager/c_api.h"
#include "tensorflow/c/eager/c_api_experimental.h"
#include "tensorflow/c/eager/parallel_device/parallel_device_testlib.h"
#include "tensorflow/c/eager/tfe_context_internal.h"

                            const char* const* underlying_devices,
                            int num_underlying_devices,
                            TFE_CustomDevice* device, void** device_info) {
  device->copy_tensor_to_device = &CopyToParallelDevice;
  device->copy_tensor_from_device = &CopyTensorFromParallelDevice;
  device->delete_device = &DeleteParallelDevice;
  device->execute = &ParallelDeviceExecute;

  const char* device_name = "/job:localhost/replica:0/task:0/device:CUSTOM:0";
  const char* first_device_name =
      "/job:localhost/replica:0/task:0/device:CPU:0";
  const char* second_device_name =
      "/job:localhost/replica:0/task:0/device:CPU:1";
  std::array<const char*, 2> underlying_devices{first_device_name,
                                                second_device_name};

                                  absl::optional<int64_t> step_id) const {
  for (int device_index = 0; device_index < underlying_devices_.size();
       ++device_index) {
    DeviceThread* device_thread = device_threads_[device_index].get();
    std::vector<TFE_TensorHandle*> device_inputs;
    device_inputs.reserve(inputs.size());
    for (int input_index = 0; input_index < inputs.size(); ++input_index) {

// Create and modify a variable placed on a parallel device which composes
// `first_device` and `second_device`.
void BasicTestsForTwoDevices(TFE_Context* context, const char* first_device,
                             const char* second_device) {
  // Register the custom device
  std::unique_ptr<TF_Status, decltype(&TF_DeleteStatus)> status(
      TF_NewStatus(), TF_DeleteStatus);