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

CUDA performance regression of n-body after LLVM15 upgrade #6957

Closed
turbo0628 opened this issue Dec 22, 2022 · 2 comments · Fixed by #7012
Closed

CUDA performance regression of n-body after LLVM15 upgrade #6957

turbo0628 opened this issue Dec 22, 2022 · 2 comments · Fixed by #7012
Assignees
@turbo0628

Comments

@turbo0628
Copy link
Member

Charts:

Compare with previous releases
image

Perfmon records
image
@turbo0628 turbo0628 self-assigned this Dec 22, 2022
@taichi-gardener taichi-gardener moved this to Untriaged in Taichi Lang Dec 22, 2022
@turbo0628 turbo0628 mentioned this issue Dec 22, 2022
Closed
2 tasks
@turbo0628
Copy link
Member Author

turbo0628 commented Dec 29, 2022 •
edited
Loading

After checking the nvptx code, the most obvious thing is that ftz option is not working for LLVM15

Do we need to change the options here?

@turbo0628
Copy link
Member Author

turbo0628 commented Dec 29, 2022 •
edited
Loading

Also, the rsqrt is replaced by slower sqrt + rn combo:

image

need to fix this.

Update: We have no change in the rsqrt implementation. The problem is LLVM15 refuses to risk data precision with the rsqrt.approx instruction. Miss some llvm math options?

@feisuzhu feisuzhu moved this from Untriaged to In Progress in Taichi Lang Dec 30, 2022
@turbo0628 turbo0628 mentioned this issue Dec 30, 2022
Merged
bobcao3 pushed a commit that referenced this issue Dec 30, 2022
@turbo0628
[cuda] Fix LLVM15 rsqrt perf regression (#7012)
b8dec02 
Fixes #6957
@github-project-automation github-project-automation bot moved this from In Progress to Done in Taichi Lang Dec 30, 2022
feisuzhu pushed a commit to feisuzhu/taichi that referenced this issue Jan 5, 2023
@turbo0628 @feisuzhu
[cuda] Fix LLVM15 rsqrt perf regression (taichi-dev#7012)
7882ed1 
Fixes taichi-dev#6957
quadpixels pushed a commit to quadpixels/taichi that referenced this issue May 13, 2023
@turbo0628 @quadpixels
[cuda] Fix LLVM15 rsqrt perf regression (taichi-dev#7012)
39e4581 
Fixes taichi-dev#6957
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@turbo0628 turbo0628

Labels
None yet
Projects
Taichi Lang
Status: Done
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

[cuda] Fix LLVM15 rsqrt perf regression turbo0628/taichi
1 participant
@turbo0628