Skip to content

[Feature]Cutlass (re) integration#1393

Closed
abhilash1910 wants to merge 2 commits intobitsandbytes-foundation:mainfrom
abhilash1910:cutlass
Closed

[Feature]Cutlass (re) integration#1393
abhilash1910 wants to merge 2 commits intobitsandbytes-foundation:mainfrom
abhilash1910:cutlass

Conversation

@abhilash1910
Copy link
Copy Markdown
Contributor

Hi @TimDettmers , @Titus-von-Koeller & @matthewdouglas ,
This is part of our initial discussion to (re) integrate and try out cutlass integration support for gemm device kernel.
Currently this is a WIP as I am testing out standalone cutlass gemms (without cute) and would incorporate cute if required for performance. Yes this took some time as was caught up in some other work , but plan to actively add/integrate this feature to cuda backend. In longer run , we plan to integrate this in sycl backend as well.
cc @mehdi-goli for review and feedbacks .

@abhilash1910 abhilash1910 marked this pull request as draft October 16, 2024 16:09
@lamikr
Copy link
Copy Markdown

lamikr commented Jan 27, 2025

Will this be behind some kind of build layer so that it can be disabled for example from AMD's rocm build target which does not have direct cutlass replacement available. In rocm case, composable kernel could perhaps be used for providing a similar kinf of features what cutlass have, but it would need some porting work.

@matthewdouglas
Copy link
Copy Markdown
Member

Closing as this is stale and quite out of date.

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