Skip to content

Conversation

@Dasor
Copy link

@Dasor Dasor commented Sep 9, 2025

Really simple change to emit linalg.batch_matmul instead of linalg.matmul if the input to tt.dot is a 3d tensor.

@ziliangzl
Copy link
Contributor

Hi @Dasor, I'm just curious if there would be a Triton kernel that could actually utilize a 3d tt.dot?

@Dasor
Copy link
Author

Dasor commented Sep 10, 2025

Hi @ziliangzl yes there are, I added support for it because this test from triton-cpu was failing:

https://github.com/triton-lang/triton-cpu/blob/main/python/test/unit/language/test_core.py#L4203

@ziliangzl
Copy link
Contributor

Oh I see, I honestly didn’t know that a 3D tt.dot could be implemented this way.
Thanks a lot for pointing me to the test, that’s really helpful!🙏

@Dasor
Copy link
Author

Dasor commented Sep 15, 2025

May I have this PR and my other one review? @red1bluelost

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.

2 participants