[TF:TRT] Support GatherV2 op when indices is constant

[meena-at-work]

Convert the 'indices' input to a GatherV2 op into a constant layer when 'indices' is a constant.

Modify the unit tests to generate constant indices.

Signed-off-by: Meenakshi Venkataraman meenakshiv@nvidia.com

Tagging @bixia1 for review
CC: @tfeher @DEKHTIARJonathan

[google-cla]

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 with @googlebot 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 Google project maintainer to go/cla#troubleshoot (Public version).
• 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.

ℹ️ Googlers: Go here for more info.

[meena-at-work]

@googlebot I signed it!

[bixia1]

In your PR description, you have:
Note: Support for this op in TRT implicit batch mode when both input and indices are both constants is limited to the case when batch size is 1.

we actually require batch size 1 when (1) both input and indices are tensors, or (2) both input and indices are constants.
But why do we add the case for (2), as we apply constant folding? In TF-TRT bridge, we report a problem when we see constant operations like this.

[meena-at-work]

@bixia1 , regarding your comment:

we actually require batch size 1 when (1) both input and indices are tensors, or (2) both input and indices are constants.
But why do we add the case for (2), as we apply constant folding? In TF-TRT bridge, we report a problem when we see constant operations like this.

Thanks for pointing this out.

1. is already covered in the converter -- and we flag it.
2. I added support for with this patch. Previously we weren't supporting indices as constant at all, now we do.

I could remove my note on the PR, as it sounds redundant.

[bixia1]

Looks good to me. Thanks for your work!

[bixia1]

Looks good to me. I modified the PR description, please check.
Would you please squash your commits into one before I approve it? This is because the commit history will become part of the commit message if you don't squash the commits.

[meena-at-work]

@bixia1 -- squash done.

[bixia1]

I am working on fixing some minor format issues and merging this PR from my end.