Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Enable geometric ops, exp2, expm1, rsqrt & erfc for BFloat16 on CUDA #57913

Closed
wants to merge 15 commits into from
Closed

Enable geometric ops, exp2, expm1, rsqrt & erfc for BFloat16 on CUDA #57913

wants to merge 15 commits into from

Conversation

imaginary-person
Copy link
Contributor

@imaginary-person imaginary-person commented May 9, 2021 •
edited

Ops enabled for BFloat16 on CUDA (12 in total):

acos
asin
atan
cosh
sin
sinh
tan
sinc
exp2
erfc
expm1
rsqrt

Enabled backward for cos on CUDA.

@facebook-github-bot
Copy link
Contributor

facebook-github-bot commented May 9, 2021 •
edited

💊 CI failures summary and remediations

As of commit 6747c91 (more details on the Dr. CI page):

  • 1/1 failures possibly* introduced in this PR
    • 1/1 non-scanned failure(s)
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 to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

@imaginary-person imaginary-person changed the title Enable geometric ops for BFloat16 on CUDA Enable geometric ops, exp2 & erfc for BFloat16 on CUDA May 9, 2021
@imaginary-person imaginary-person changed the title Enable geometric ops, exp2 & erfc for BFloat16 on CUDA Enable geometric ops, exp2, expm1, rsqrt, nan_to_num & erfc for BFloat16 on CUDA May 9, 2021
@imaginary-person
Copy link
Contributor Author

TestUnaryUfuncsCUDA.test_reference_numerics_extremal_nan_to_num_cuda_bfloat16 fails with the following error:

AssertionError: False is not true : Tensors failed to compare as equal!
With rtol=0.016 and atol=1e-05, found 2 element(s) (out of 812) 
whose difference(s) exceeded the margin of error (including 0 nan comparisons). 
The greatest difference was inf (3.3895313892515355e+38 vs. inf), 
which occurred at index 0.

@imaginary-person imaginary-person changed the title Enable geometric ops, exp2, expm1, rsqrt, nan_to_num & erfc for BFloat16 on CUDA Enable geometric ops, exp2, expm1, rsqrt & erfc for BFloat16 on CUDA May 10, 2021
@imaginary-person imaginary-person marked this pull request as ready for review May 10, 2021 20:05
@imaginary-person imaginary-person mentioned this pull request May 10, 2021
Closed
@bdhirsh bdhirsh requested a review from ngimel May 10, 2021 22:01
@bdhirsh bdhirsh added the triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module label May 10, 2021
ngimel
ngimel approved these changes May 11, 2021
View reviewed changes
Copy link
Collaborator

@ngimel ngimel left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM if CI passes

@facebook-github-bot
Copy link
Contributor

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

@codecov
Copy link

codecov bot commented May 11, 2021

Codecov Report

Merging #57913 (6747c91) into master (0c2d382) will increase coverage by 0.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master   #57913      +/-   ##
==========================================
+ Coverage   76.82%   76.83%   +0.01%     
==========================================
  Files        1985     1985              
  Lines      197151   197151              
==========================================
+ Hits       151454   151479      +25     
+ Misses      45697    45672      -25

@facebook-github-bot
Copy link
Contributor

@ngimel merged this pull request in fa31891.

krshrimali pushed a commit to krshrimali/pytorch that referenced this pull request May 19, 2021
@imaginary-person
Enable geometric ops, exp2, expm1, rsqrt & erfc for BFloat16 on CUDA (p…
1b53a49 
…ytorch#57913)

Summary:
Ops enabled for BFloat16 on CUDA (12 in total):

`acos`
`asin`
`atan`
`cosh`
`sin`
`sinh`
`tan`
`sinc`
`exp2`
`erfc`
`expm1`
`rsqrt`

Enabled backward for `cos` on CUDA.

Pull Request resolved: pytorch#57913

Reviewed By: mruberry

Differential Revision: D28342969

Pulled By: ngimel

fbshipit-source-id: 3c140fe408cbf93b21296a52d95ef0a0ccd96503
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@ngimel ngimel ngimel approved these changes

Assignees
No one assigned
Labels
cla signed Merged open source triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
5 participants
@imaginary-person @facebook-github-bot @ngimel @bdhirsh @pytorchbot