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.

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

[CINN] Accelerate recompute ips #70386

Merged
merged 2 commits into from
Dec 23, 2024
Merged

Conversation

chen2016013
Copy link
Contributor

PR Category

CINN

PR Types

Bug fixes

Description

pcard-67164
some benchmark models show worse performance on ips after turning on Auto-ecompute Flag. One of the reasons we've found is that SUM op which is recomputable actually will be changed into compute intensive REDUCE_SUM op and takes long time to run. Therefore we remove sum op from default recomputable ops.

Copy link

paddle-bot bot commented Dec 20, 2024

你的PR提交成功,感谢你对开源项目的贡献!
请关注后续CI自动化测试结果,详情请参考Paddle-CI手册
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

"pd_op.log",
"pd_op.log1p",
"pd_op.logit",
"pd_op.max",
# "pd_op.max",
# "pd_op.min",
Copy link
Contributor

Choose a reason for hiding this comment

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

pd_op.any是否也要过滤掉?

Copy link
Contributor

Choose a reason for hiding this comment

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

slice, gather, transpose这些算子是否做recompute有哪些判定条件?直观上看这些算子recompute有可能会引起显存增加或者性能下降的情况

Copy link
Contributor Author

Choose a reason for hiding this comment

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

没有特殊判定条件。后续在分析一下,确认有问题的话(连同下面说的any)一起ban掉

Copy link
Contributor

@zhangbo9674 zhangbo9674 left a comment

Choose a reason for hiding this comment

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

LGTM

@chen2016013 chen2016013 merged commit 5994219 into PaddlePaddle:develop Dec 23, 2024
28 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants