[GitHub] [tvm] echuraev commented on pull request #8313: [Metal] Add pass for splitting kernel with huge number of args

[GitBox <gi...@apache.org>]

echuraev commented on pull request #8313:
URL: https://github.com/apache/tvm/pull/8313#issuecomment-870453510


   > Not exactly, but I've dealt with a similar issue. My mitigation was to limit the maximum fusion depth, which breaks large parameter kernels into smaller ones. But that is not guaranteed to work and not predictable. I can imagine that having a pass like this that allows more fine-grained controls might be necessary in some cases.
   > 
   > @echuraev FYI you can cap the fuse depth by
   > 
   > https://github.com/apache/tvm/blob/720e7b1ebd9b789a1100dee7536d0633c7941dd1/tests/python/relay/test_pass_fuse_ops.py#L755
   
   Thank you! I thought about the reducing fuse depth, but as you mentioned, it is not predictable and not guaranteed to work. This is why I think that this approach with splitting kernels is more robust.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org