Should we only update the `_u` in training time?

[watsonyanghx]

In https://github.com/pfnet-research/sngan_projection/blob/master/source/links/sn_convolution_2d.py#L79, the self.u is updated only in training time, while in https://github.com/pfnet-research/sngan_projection/blob/master/source/links/sn_linear.py#L66 and https://github.com/pfnet-research/chainer-gan-lib/blob/master/common/sn/sn_convolution_2d.py#L77, the self.u is updated both at training and testing time.

So, which code is right?