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.
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
Disables denormal floating numbers on ARM CPU #115184
Disables denormal floating numbers on ARM CPU #115184
Conversation
->Supports torch.set_flush_denormal(bool) ->Datatypes supported: FP64, FP32, BFloat16 Signed-off-by: maajidkhann <maajidkhan.n@fujitsu.com>
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/115184
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 8b47964 with merge base daf89b4 (): This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Hello @jgong5 . Thanks for reviewing and approving the changes. Can you please trigger the CI Pipelines from your end, so that we can get these changes merged into main while other folks review the changes. |
Thank you for running the CI pipelines. Looks like all the pipelines have passed with no failures. Can this now be merged if already been reviewed. This would be of great help. |
Hello @jgong5 Jiong. Can you help us merge this pull request or if it needs additional review, can you please drop a note to the maintainer to have a look at this change and help merging it. Thanks. |
@pytorchbot merge |
Merge failedReason: This PR needs a If not, please add the To add a label, you can comment to pytorchbot, for example For more information, see Details for Dev Infra teamRaised by workflow job |
@pytorchbot merge |
Merge startedYour 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 |
*set_flush_denormal() is now supported on ARM CPU's. PR:pytorch#115184 Signed-off-by: maajidkhann <maajidkhan.n@fujitsu.com>
*set_flush_denormal() is now supported on ARM CPU's. PR:pytorch#115184 Signed-off-by: maajidkhann <maajidkhan.n@fujitsu.com>
**Documentation update for set_flush_denormal():** -> set_flush_denormal() is now supported on ARM CPU's. -> **PR:** #115184 (Already merged) **Reference page:** https://pytorch.org/docs/stable/generated/torch.set_flush_denormal.html Pull Request resolved: #119354 Approved by: https://github.com/drisspg
**Documentation update for set_flush_denormal():** -> set_flush_denormal() is now supported on ARM CPU's. -> **PR:** #115184 (Already merged) **Reference page:** https://pytorch.org/docs/stable/generated/torch.set_flush_denormal.html Pull Request resolved: #119354 Approved by: https://github.com/drisspg
Motivation:
Denormal numbers are used to store extremely small numbers that are close to 0. Denormal numbers can incur extra computational cost. To solve the low performance issue caused by denormal numbers, Pytorch supports flushing denormal numbers and it successfully configures flush denormal mode
Currently set_flush_denormal() is only supported on x86 architectures supporting SSE3 (https://pytorch.org/docs/stable/generated/torch.set_flush_denormal.html (Opens in new window or tab) and now we want to extend this functionality for ARM architecture.
This PR:
->Supports set_flush_denormal() on ARM.
->Datatypes supported and tested: FP64, FP32, BFloat16
cc @jgong5 @mingfeima @XiaobingSuper @sanchitintel @ashokei @jingxu10