guard against None in pytorch get_xla_supported_devices

[ckchow]

What does this PR do?

Guard against None when checking if TPU's are available. This happens (e.g.) if you use google's standard deep learning containers on CPU only machines.

Fixes #9552

Does your PR introduce any breaking changes? If yes, please list them.

None

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

Make sure you had fun coding 🙃

[ckchow]

I'm actually not sure how to write a proper test for this. Seems like it would have to pull an XLA container.

[kaushikb11]

I'm actually not sure how to write a proper test for this. Seems like it would have to pull an XLA container.

We have TPU CI tests in place, But for this, we would require to pull an XLA container with no TPUs attached. Could you please add a comment for this in the code?

Do you think "the guard against None and return an empty list" should be added to the XLA's API itself? cc: @JackCaoG

[SeanNaren]

Thanks for this! At the least I think add a todo saying that this needs to be tested in a XLA supported container without TPUs!

[kaushikb11]

Blocking this merge for now. I have asked the XLA team if this change could instead be introduced on the pt/xla's side

[JackCaoG]

@kaushikb11 so get_xla_supported_devices will never return a empty list, so maybe we can just check if get_xla_supported_devices('TPU') != None

[ckchow]

I feel like somebody might change get_xla_supported_devices to return empty lists in the future (i.e. if I were a contributor to that codebase I would feel tempted to do so), so I think checking for None and len() > 0 explicitly is safer. I guess it's also correct to check tpus != None and tpus, but I always forget that empty lists are false in python so that feels a little unexpected.

[JackCaoG]

I could try to change the pt/xla code to return an empty list, but

1. That will also require me to change a couple other places in pt/xla
2. This is BC breaking, if there are users/frameworks that explicitly checking for !=None that will also have to change.

so I will prefer to leave the return None if no device found logic

[tchaton]

LGTM !

[kaushikb11]

Thanks @ckchow for the contribution! We always look forward to improving our compatibility with TPUs! :)

[codecov]

Codecov Report

Merging #9572 (a3e9139) into master (b530b7a) will decrease coverage by 4%.
The diff coverage is 0%.

@@           Coverage Diff           @@
##           master   #9572    +/-   ##
=======================================
- Coverage      93%     89%    -4%     
=======================================
  Files         178     178            
  Lines       15652   15650     -2     
=======================================
- Hits        14508   13899   -609     
- Misses       1144    1751   +607