Tile Random Tensor Without Loosing Randomness in TensorFlow

In the following example I have a random row tensor r (which for the sake of the example is just a simple tf.random_normal but it could be a complicated graph instead whose value is random).

It seems like that in TensorFlow the tf.tile operation is run only after the value of r is computed, indeed all the rows of the tiled tensor t are the same. Is there a way to tile a tensor without loosing the randomness of each row?

import tensorflow as tf

r = tf.random_normal(shape=[1, 2])

t = tf.tile(r, tf.pack([3, 1]), name=None)

session = tf.Session()

print(session.run(t))

session.close()

Result:

[[ 0.29335016  0.11843499]
 [ 0.29335016  0.11843499]
 [ 0.29335016  0.11843499]]