Update fold_old_batch_norms.cc to accommodate 'NCHW' format.

[joeyearsley]

Fixes the problem of using fused batch normalization and this transform, only shows up when using 'NCHW' as the default is 'NHWC'.

[googlebot]

Thanks for your pull request. It looks like this may be your first contribution to a Google open source project (if not, look below for help). Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

📝 Please visit https://cla.developers.google.com/ to sign.

Once you've signed (or fixed any issues), please reply here (e.g. I signed it!) and we'll verify it.

---

What to do if you already signed the CLA

Individual signers

• It's possible we don't have your GitHub username or you're using a different email address on your commit. Check your existing CLA data and verify that your email is set on your git commits.

Corporate signers

• Your company has a Point of Contact who decides which employees are authorized to participate. Ask your POC to be added to the group of authorized contributors. If you don't know who your Point of Contact is, direct the project maintainer to go/cla#troubleshoot.
• The email used to register you as an authorized contributor must be the email used for the Git commit. Check your existing CLA data and verify that your email is set on your git commits.
• The email used to register you as an authorized contributor must also be attached to your GitHub account.

[joeyearsley]

I signed it!

[googlebot]

CLAs look good, thanks!

[mingxingtan]

This would fail the whole program if the conv_node doesn't have "data_format" attribute. You probably need to first verify it has the attribute "data_format" before calling CopyNodeAttr?

[joeyearsley]

I don't see any checks like this in any other part of the code, please could you direct me to some.

Also, since it is now part of the api in tf.nn.conv2d does it not get this attribute regardless? - I'm not sure but I thought that it would fallback to the default.

[mingxingtan]

This is just an extra safety check for legacy tf graphs. You are right, If the original graph is created with tf.nn.conv2d, then this check is not needed; but I am not whether there are any legacy tf graphs that do not contain this attr.

[joeyearsley]

@mingxingtan Added the check as requested.

[martinwicke]

@mingxingtan can you take another look?

[mingxingtan]

I don't see where is HasAttr defined. Maybe you can directly use something like:

if (!conv_node.attr().count("data_format")) {
  CopyNodeAttr(conv_node, "data_format", "data_format", &bias_add_node);
}

[joeyearsley]

Updated as requested