System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): yes
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 16.04
• TensorFlow installed from (source or binary): binary
• TensorFlow version (use command below): v1.4.0-rc1-11-g130a514 1.4.0
• Python version: 3.5
• Bazel version (if compiling from source):
• GCC/Compiler version (if compiling from source):
• CUDA/cuDNN version: CUDA 8.0 / cuDNN 6
• GPU model and memory:
• Exact command to reproduce:

tf.depth_to_space and tf.space_to_depth support data_format='NCHW' on GPU. However, _SpaceToDepthGrad and _DepthToSpaceGrad are not aware of data_format. Maybe they would need to propagate op.get_attr('data_format').

Source code / logs

import tensorflow as tf
x = tf.zeros([1, 4, 1, 1])
y = tf.depth_to_space(x, 2, data_format='NCHW')
# ValueError: Dimension size must be evenly divisible by 2 but is 1 for 'gradients ...
g = tf.gradients(y, x)