masahi commented on pull request #9482: URL: https://github.com/apache/tvm/pull/9482#issuecomment-1031864766
I would check if `vta.build(...)` is really honoring `disabled_passes=[...]` when it runs. You can build VTA and run this test locally. Also, what do you think about running the TIR CSE pass optional? I have a feeling that it might bring some disruption if enabled globally, without much benefit for standard backends such as LLVM and CUDA which, until today, didn't see a need for TIR-level CSE (the example I gave, CSE across CPU & GPU, is the only good use case I'm aware of). That will also remove the need to fix existing tests. One way is to make TIR-CSE run only when `opt-level=4`, but I hope there is a better way to enable certain TIR passes. -- 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]
