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

[TVM][LANG] Add eager simplification for operations with FloatImm #2615

Merged
merged 6 commits into from
Feb 19, 2019
Merged

[TVM][LANG] Add eager simplification for operations with FloatImm #2615

merged 6 commits into from
Feb 19, 2019

Conversation

icemelon
Copy link
Member

Add eager simplification for operations of add, sub, mul, div, ceil, floor, round, trunc, and cast with float constants.

@icemelon icemelon requested a review from tqchen February 18, 2019 00:12
@tqchen
Copy link
Member

tqchen commented Feb 18, 2019

Thank you @icemelon9 can you add some test cases as per https://docs.tvm.ai/contribute/code_review.html#ensure-test-coverage

@tqchen tqchen self-assigned this Feb 18, 2019
@tqchen tqchen added the status: need update need update based on feedbacks label Feb 18, 2019
@icemelon
Copy link
Member Author

@tqchen Test case for eager simplification is added.

@icemelon icemelon mentioned this pull request Feb 18, 2019
Merged
@icemelon
Copy link
Member Author

Thanks @tqchen for the review.

@icemelon icemelon added status: accepted and removed status: need update need update based on feedbacks labels Feb 19, 2019
@icemelon icemelon merged commit c59a78e into apache:master Feb 19, 2019
@icemelon icemelon deleted the simplify branch February 19, 2019 17:15
wweic pushed a commit to neo-ai/tvm that referenced this pull request Feb 20, 2019
@icemelon
[TVM][LANG] Add eager simplification for operations with FloatImm (ap…
69a8fd9 
…ache#2615)

* Add eager simplication for FloatImm

* fix

* fix lint

* Fix gcc warning

* fix

* Add test case
@yzhliu yzhliu mentioned this pull request Mar 2, 2019
Closed
28 tasks
@merrymercy
Copy link
Member

merrymercy commented Apr 9, 2019 •
edited
Loading

This PR introduced a performance regression on resnet-18 on Mali GPUs.
Some loops cannot be vectorized after this PR. I am not sure about the reasons, will investigate later (volunteers are welcomed).

cc @yzhliu @icemelon9

@yzhliu
Copy link
Member

yzhliu commented Apr 11, 2019

Rough investigation, regression disappears by applying this patch: https://gist.github.com/yzhliu/0d2cdfc9fa92127b81a1298d5bec55a0

yzhliu added a commit to yzhliu/tvm-1 that referenced this pull request Apr 11, 2019
@yzhliu
revert part of apache#2615. mali performance can be reproduced
62fd8e9
@icemelon
Copy link
Member Author

This may break the arange op because it relies on the simplification on floating expr to calculate the output shape. Maybe the new canonical simplification can handle this instead of eager simplification. Could you test it?

@merrymercy merrymercy mentioned this pull request Apr 24, 2019
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@tqchen tqchen tqchen approved these changes

Assignees

@tqchen tqchen

Labels
status: accepted
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@icemelon @tqchen @merrymercy @yzhliu