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

[MetaSchedule] Skip custom_rule in case of None #14732

Open
wants to merge 2 commits into
base: main
Choose a base branch
from

Conversation

apeskov
Copy link
Contributor

@apeskov apeskov commented Apr 27, 2023

Custom rule mechanics allows to have "schedule_rule" attribute equals "None". This value will be ignored by ApplyCustomRuleNode. But also this block will be ignore by all other rules. It happens because of improper check in PostOrderApplyNode.

This patch allows to use regular rules in case of schedule_rule=="None".

Signed-off-by: Alexander Peskov <peskovnn@gmail.com>
@tvm-bot
Copy link
Collaborator

tvm-bot commented Apr 27, 2023

Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.

Generated by tvm-bot

@Hzfengsy
Copy link
Member

cc @junrushao

Signed-off-by: Alexander Peskov <peskovnn@gmail.com>
@apeskov
Copy link
Contributor Author

apeskov commented Apr 28, 2023

@junrushao could you please take a look. In your previous patch #13195 with refactoring of custom rules you already had a discussion with @masahi about auto tensorization issue.

But there is an another aspect. In case of schedule_rule == "None", ApplyCustomRule and sch->Unannotate will not be called. So block will not be processed by any schedule rule never. I'm not sure if it was in the design.

@junrushao
Copy link
Member

I vaguely recall that there is a reason that we use custom_rule: None, which means we wanted to design space generator to intentionally skip a block. My assumption was that if we have a good usecase if we want the generator not to skip, then perhaps we could unannotate the block first - but I'm happy to change my mind if the situation above has been taken good consideration

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.

4 participants