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

[QST] Are there plans to add specialisations for Sm90? #1123

Open
joerowell opened this issue Oct 4, 2023 · 8 comments
Open

[QST] Are there plans to add specialisations for Sm90? #1123

joerowell opened this issue Oct 4, 2023 · 8 comments
Labels
help wanted Interested in support from contributors inactive-30d question Question
Milestone

Comments

@joerowell
Copy link
Contributor

What is your question?
I recently tried to change the type tags on the DGEMM examples to cutlass::arch::Sm90, which caused a load of compile errors. This is primarily because there's certain template specialisations missing for Sm90, e.g there's no specialisation like this.

Is this something that's on the plans to be added / are PRs welcome?

@hwu36
Copy link
Collaborator

hwu36 commented Oct 4, 2023

the link points to dual gemm example not dgemm. do you want dual gemm or dgemm?

@joerowell
Copy link
Contributor Author

joerowell commented Oct 4, 2023 via email

@hwu36
Copy link
Collaborator

hwu36 commented Oct 4, 2023

@danthe3rd

you should be able to still run this example on h100 without changing the source code. just use sm90 in nvcc/cmake command line.

this example does not use new features of hopper. we are looking to write a hopper dedicated version using cutlass 3.x

@vmarkovtsev
Copy link
Contributor

vmarkovtsev commented Oct 5, 2023

Note: we with @joerowell and others are working on a diverged hard fork of xformers that has little to do with the mainstream 😉 We are rewriting the orignal kernels, doing a lot of optimizations, fixing bugs, etc.

@mnicely
Copy link
Collaborator

mnicely commented Oct 10, 2023

@joerowell optimizations for dual GEMM are on our roadmap for H1'24.

@mnicely mnicely added the help wanted Interested in support from contributors label Oct 10, 2023
@mnicely mnicely added this to the Backlog milestone Oct 19, 2023
Copy link

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

Copy link

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

@pansershrek
Copy link

@mnicely Hi! Is any update in roadmap for dual GEMM optimizations? Will this optimization appear? Or you reprioritized this task? Or it is too difficult to optimize for hopper ?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
help wanted Interested in support from contributors inactive-30d question Question
Projects
None yet
Development

No branches or pull requests

5 participants