QR decomposition of type float16?

Question

https://www.tensorflow.org/api_docs/python/tf/linalg/qr

The documentation states that a tensor must be of type float64, float32, half, complex64, or complex128.

tn = tf.constant([ [1, 2], [3, 4] ], dtype=tf.float32)
[Q, R] = tf.linalg.qr(tn);

tn = tf.constant([ [1, 2], [3, 4] ], dtype=tf.float16)
[Q, R] = tf.linalg.qr(tn);

Using single precision works as intended but half precision returns the following error:

Could not find device for node: {{node Qr}} = Qr[T=DT_HALF, full_matrices=false