Siyuan Feng
Siyuan Feng
Sorry for the late response. Please rebase the code, and we can get it in
cc TVMC maintainer @leandron
Thanks for pinging me. However, I'm not familiar with Windows and WSL. I'm afraid that I'm not able to review this one
Sorry for the late response, and thanks for the great catch. One thing is that did you test if the TIR function itself is i64-based? It would be good if...
I prefer to annotate non-restrict buffers as it's unusual. The syntax can be: ```python @T.prim_func def buggy_kernel(x: T.Tensor[(num_tokens, ), 'float', restrict=False], y: T.Tensor[(num_tokens, ), 'float', restrict=False]): with T.Kernel(num_tokens, threads=32) as...
I will help with the rebase and integration process
I wonder why runtime is enough, as tilelang reuses TVM's compiation and codegen utils.
As 3.8 is deprecated, `from __future__ import annotations` is no longer needed