f=tpu_subgraph,
      device_ordinal=[0])

static_shape = tf.placeholder(tf.int32, shape=([3]), name='static_size')

w = tf.Variable(tf.constant([[1.0], [2.0], [3.0]]), name='w')

w_dyn = tf.SetDynamicDimensionBounds(w, static_size])
tpu_call([w_dyn])
}];
  let arguments = (ins
    TF_Tensor:$input,
    TF_I32OrI64Tensor:$static_shape
  );

  const int state_size =
      reduce_dataset->getAttrOfType<ArrayAttr>("Tstate").size();
  for (int i = 0; i < state_size; ++i) {
    body_returns.push_back(dataset_if.getResult(i));
  }
  // Copy the arguments but skip the states and the loop condition
  // which are updated in while body.
  for (int i = state_size + 1; i < body_args.size(); ++i) {

      return rewriter.notifyMatchFailure(op, "missing resource subtype");
    }

    int counter_size = tensorflow::GetCounterSize(alg);
    int state_size = counter_size + tensorflow::RNG_KEY_SIZE;
    RankedTensorType res_type =
        RankedTensorType::get({state_size}, state_element_type);
    if (res_type != GetResourceSubtype(rng_op.getResource())) {
      return rewriter.notifyMatchFailure(op, "unexpected resource subtype");

	STATX_MNT_ID_UNIQUE                         = 0x4000
	STATX_MODE                                  = 0x2
	STATX_MTIME                                 = 0x40
	STATX_NLINK                                 = 0x4
	STATX_SIZE                                  = 0x200
	STATX_TYPE                                  = 0x1
	STATX_UID                                   = 0x8
	STATX__RESERVED                             = 0x80000000

                                                  : input_type.getDimSize(1);
    const int state_size = cell_state.getNumElements();

    // Check if the dimension of the inputs matches.
    if ((output_state_size != n_batch * n_output) ||
        (state_size != n_batch * n_cell) ||
        (input_to_output_weights.getDimSize(1) != n_input) ||

        (SavedModelEstimator).
    *   Adds leaf index modes as an argument.
    *   Allow a different output shape from the input in
        tf.contrib.image.transform.
    *   Change the state_size order of the StackedRNNCell to be natural order.
        To keep the existing behavior, user can add reverse_state_order=True
        when constructing the StackedRNNCells.