yelite commented on PR #14665:
URL: https://github.com/apache/tvm/pull/14665#issuecomment-1526860017

   @areusch Just want to clarify, I don't think the approach I shared is 
strictly better than the one implemented in this PR. I just wanted to share 
some thought on this, because I faced the same problem when benchmarking TVM on 
some models. I think which solution is better really depends on the use case. 
If it's for error reporting at some point of the transformation pipeline, being 
confined in IR graph is a much better solution.
   
   > there's no forcing function to ensure that passes which run after this one 
also adjust the tracking information placed in the IR
   
   Just for my own understanding, how does the MultiSpan approach ensure the 
span being passed along during transformation? Does it rely on the pass to copy 
span when creating a new function out of the old one?


-- 
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: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to