CUDA BF16 norm

[zasdfgbnm]

Fixes #{issue number}

[dr-ci]

💊 CI failures summary and remediations

As of commit 219512e (more details on the Dr. CI page):

---

💚 💚 Looks good so far! There are no failures yet. 💚 💚

---

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group.

See how this bot performed.

This comment has been revised 3 times.

[codecov]

Codecov Report

Merging #48806 (219512e) into master (1eed54d) will decrease coverage by 0.00%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master   #48806      +/-   ##
==========================================
- Coverage   80.79%   80.79%   -0.01%     
==========================================
  Files        1865     1865              
  Lines      201099   201099              
==========================================
- Hits       162474   162468       -6     
- Misses      38625    38631       +6     

[ngimel]

cc @mruberry @kurtamohler for norm testing. From my (admittedly very cursory) look there's no test where norm would be tested for a cross-product of dtypes and p's. I also don't have a good idea of how torch.norm and torch.linalg.norm is tested. With that said, I'm going to land this PR as it does not make things worse, and it adds tests for bfloat16 and p=2

[facebook-github-bot]

@ngimel has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[mruberry]

cc @mruberry @kurtamohler for norm testing. From my (admittedly very cursory) look there's no test where norm would be tested for a cross-product of dtypes and p's. I also don't have a good idea of how torch.norm and torch.linalg.norm is tested. With that said, I'm going to land this PR as it does not make things worse, and it adds tests for bfloat16 and p=2

Thanks, and thank you for the fyi, @ngimel!

@kurtamohler maybe we need to add another test or extend an existing test to enumerate over the dtype kwarg for torch.linalg.norm?

[kurtamohler]

Sure, I can extend the test_norm_dtype test. Might be best to wait for #48284 to land first, since it changes that test a fair bit

[facebook-github-bot]

@ngimel merged this pull request in a39398b.

[facebook-github-bot]

@ngimel merged this pull request in a39398b.