Skip to content

[cuBLAS] Add an option to disable reduced precision reductions for BF16 GEMM#89172

Closed
eqy wants to merge 3 commits into
pytorch:masterfrom
eqy:cublas_reduced_precision_bf16
Closed

[cuBLAS] Add an option to disable reduced precision reductions for BF16 GEMM#89172
eqy wants to merge 3 commits into
pytorch:masterfrom
eqy:cublas_reduced_precision_bf16

Conversation

@eqy

@eqy eqy commented Nov 17, 2022

Copy link
Copy Markdown
Collaborator

Essentially the same change as #67946, except that the default is to disallow reduced precision reductions in BFloat16 GEMMs (for now). If performance is severely regressed, we can change the default, but this option appears to be necessary to pass some addmm BFloat16 tests on H100.

CC @ptrblck @ngimel

@pytorch-bot

pytorch-bot Bot commented Nov 17, 2022 •
edited
Loading

Copy link
Copy Markdown

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/89172

Note: Links to docs will display an error until the docs builds have been completed.

⏳ No Failures, 1 Pending

As of commit 909a989:
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

@eqy eqy added the ciflow/trunk Trigger trunk jobs on your pull request label Nov 17, 2022
@soulitzer soulitzer requested a review from ngimel November 18, 2022 06:57
@soulitzer soulitzer added the triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module label Nov 18, 2022
@eqy eqy changed the title [cuBLAS] Add an option to disable reduced precision reductions for FP16 GEMM [cuBLAS] Add an option to disable reduced precision reductions for 16 GEMM Nov 19, 2022
@eqy eqy changed the title [cuBLAS] Add an option to disable reduced precision reductions for 16 GEMM [cuBLAS] Add an option to disable reduced precision reductions for BF16 GEMM Nov 19, 2022
@eqy

eqy commented Nov 22, 2022

Copy link
Copy Markdown
Collaborator Author

@pytorchmergebot rebase

@pytorchmergebot

pytorchmergebot commented Nov 22, 2022

Copy link
Copy Markdown
Collaborator

@pytorchbot successfully started a rebase job. Check the current status here

@pytorchmergebot

pytorchmergebot commented Nov 22, 2022

Copy link
Copy Markdown
Collaborator

Successfully rebased cublas_reduced_precision_bf16 onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout cublas_reduced_precision_bf16 && git pull --rebase)

@pytorchmergebot pytorchmergebot force-pushed the cublas_reduced_precision_bf16 branch from e64bf93 to 6133bcc Compare November 22, 2022 19:40
@eqy

eqy commented Dec 13, 2022

Copy link
Copy Markdown
Collaborator Author

@pytorchmergebot rebase

@pytorchmergebot

pytorchmergebot commented Dec 13, 2022

Copy link
Copy Markdown
Collaborator

@pytorchbot successfully started a rebase job. Check the current status here

@pytorchmergebot

pytorchmergebot commented Dec 13, 2022

Copy link
Copy Markdown
Collaborator

Successfully rebased cublas_reduced_precision_bf16 onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout cublas_reduced_precision_bf16 && git pull --rebase)

@pytorchmergebot pytorchmergebot force-pushed the cublas_reduced_precision_bf16 branch from 2d9b2e3 to 82d6337 Compare December 13, 2022 23:07
ngimel
ngimel approved these changes Dec 13, 2022
View reviewed changes

@ngimel ngimel left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Approving, but I'd defer to @ptrblck for what the default value should be, just making tests pass may not be reason enough to flip the default, as long as real workloads still converge.

@eqy

eqy commented Dec 20, 2022

Copy link
Copy Markdown
Collaborator Author

@pytorchmergebot rebase

@pytorchmergebot

pytorchmergebot commented Dec 20, 2022

Copy link
Copy Markdown
Collaborator

@pytorchbot successfully started a rebase job. Check the current status here

@pytorchmergebot

pytorchmergebot commented Dec 20, 2022

Copy link
Copy Markdown
Collaborator

Successfully rebased cublas_reduced_precision_bf16 onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout cublas_reduced_precision_bf16 && git pull --rebase)

@pytorchmergebot pytorchmergebot force-pushed the cublas_reduced_precision_bf16 branch from 81898db to 909a989 Compare December 20, 2022 22:31
@eqy

eqy commented Dec 21, 2022

Copy link
Copy Markdown
Collaborator Author

@pytorchmergebot merge

@pytorchmergebot

pytorchmergebot commented Dec 21, 2022

Copy link
Copy Markdown
Collaborator

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging
Check the merge workflow status
here

laurentdupin pushed a commit to laurentdupin/pytorch that referenced this pull request Apr 25, 2026
@eqy @pytorchmergebot
[cuBLAS] Add an option to disable reduced precision reductions for BF…
3e43e9f 
…16 GEMM (pytorch#89172)

Essentially the same change as pytorch#67946, except that the default is to disallow reduced precision reductions in `BFloat16` GEMMs (for now). If performance is severely regressed, we can change the default, but this option appears to be necessary to pass some `addmm` `BFloat16` tests on H100.

CC @ptrblck @ngimel
Pull Request resolved: pytorch#89172
Approved by: https://github.com/ngimel
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

ciflow/trunk Trigger trunk jobs on your pull request 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.

5 participants

@eqy @pytorchmergebot @ngimel @soulitzer @pytorchbot