`.predict` is available for Cloud TPUs, Cloud TPU, for all types of
        Keras models (sequential, functional and subclassing models).
    *   Automatic outside compilation is now enabled for Cloud TPUs. This allows
        `tf.summary` to be used more conveniently with Cloud TPUs.
    *   Dynamic batch sizes with DistributionStrategy and Keras are supported on
        Cloud TPUs.

  let summary = "A TPU core selector Op.";

  let description = [{
This Op produces a set of TPU cores (for warm-up) or a single TPU core
(for regular inference) to execute the TPU program on. The output is
consumed by TPUPartitionedCall.
  }];

  let arguments = (ins);

  let results = (outs
    Res<TF_Int32Tensor, [{A vector 1 or more TPU cores.}]>:$device_ordinals
  );
}