gpu_options->set_allow_growth(gpu_memory_allow_growth);

  (*config.mutable_device_count())["CPU"] = num_cpu_devices;

  // TODO(b/113217601): This is needed for EagerContext::runner_ to use a
  // threadpool, so that we avoid the possibility of running the runner_ in the
  // threadpool of GPU event mgr, as that can trigger more callbacks to be
  // scheduled on that same threadpool, causing a deadlock in cases where the

      TF_SetStatusFromIOError(status, errno, dst);
      return;
    }
  } else if (S_ISDIR(st.st_mode)) {
    TF_SetStatus(status, TF_FAILED_PRECONDITION, "target path is a directory");
    return;
  }

  // We cannot rename directories yet, so prevent this.
  if (stat(src, &st) != 0) {
    TF_SetStatusFromIOError(status, errno, src);
    return;
  } else if (S_ISDIR(st.st_mode)) {

  // There are two ways to represent compact row-major data
  // 1) nullptr indicates tensor is compact and row-majored.
  // 2) fill in the strides array as the real case for compact row-major data.
  // Here we choose option 2, since some frameworks didn't handle the strides
  // argument properly.
  dlm_tensor->dl_tensor.strides = stride_arr->data();

  dlm_tensor->dl_tensor.byte_offset =

  // TODO(srbs): It seems like this is used only for performance optimization
  // and not for correctness. The only downside of keeping this 1 seems to be
  // that the gradient accumulation is unbounded and we will never
  // aggressively aggregate accumulated gradients to recover memory.
  // Revisit and fix.
  return 1;
}

// Consumes references to the tensors in the gradient_tensors list and returns

    // Wait until all pending nodes have completed since they may have a
    // reference to default_cancellation_manager_. We ignore the status return
    // since we already have a bad status to propagate.
    TFE_ContextAsyncWait(context, await_status.get());
    // Reset the cancellation manager on a bad status. Otherwise we'll cancel
    // all future operations.
    default_cancellation_manager_ = absl::make_unique<CancellationManager>();
  }

      if (iter == input_nodes.end()) {
        // This node is not referenced in inputs. Add it to the body.
        body_nodes->push_back(node);
      } else {
        // This node is referenced in inputs. Currently, we place an
        // artificial restriction and require that when num_opers=-1, such
        // nodes must have a single output.
        if (node->num_outputs() != 1) {
          return InvalidArgument(

  TFE_ContextSetServerDef(ctx, 0, serialized.data(), serialized.size(), status);
  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  // Use large matrices so that RPCs don't return before we get a chance
  // to call TFE_DeleteContext.
  TFE_TensorHandle* h0_task0 = TestMatrixTensorHandle100x100(ctx);
  TFE_TensorHandle* h1_task0 = TestMatrixTensorHandle100x100(ctx);
  const char remote_device_name[] =

    }
    if (op_type_.empty()) {
      return errors::FailedPrecondition(
          "GraphOperation::Reset must be called before calling SetOpName.");
    }
    // TODO(b/145674566): We use Graph::NewName to get a unique name here but
    // this may not be consistent with python's naming policy.
    mutex_lock l(g_->mu);
    op_.reset(new TF_OperationDescription(g_, op_type_.c_str(),

        tensorflow::unwrap(handles[i]);
    if (tensorflow::CustomDeviceTensorHandle::classof(unwrapped_handle)) {
      // One of the inputs we're trying to pack is on a custom device. We'll let
      // the first custom device we see handle all of the packing.
      auto* custom_device_handle =
          tensorflow::down_cast<tensorflow::CustomDeviceTensorHandle*>(
              unwrapped_handle);

    return nullptr;
  }
  std::vector<int64_t> dimvec(num_dims);
  for (int i = 0; i < num_dims; ++i) {
    dimvec[i] = static_cast<int64_t>(dims[i]);
  }

  // TODO(apassos) do we need to wrap the deallocator here to make sure to sync
  // the device?
  TF_ManagedBuffer* buf =
      new TF_ManagedBuffer(data, len, deallocator, deallocator_arg,
                           /*owns_memory=*/false);