it to use the saved average mean and invstd to scale the data instead of
normalizing the batch.

with other parts of the library.

spec says it should.

cudnnAddTensor() function and updated the specs and asserts accordingly.

kernel was still running.

CUDA version of add_bias_gradient().

batch_normalize_conv.

of add to them so that it's consistent with how the layer interface expects
this to be done.