manupa-arm commented on pull request #9706: URL: https://github.com/apache/tvm/pull/9706#issuecomment-991090858
Thanks @jacobbohlin for fixing this! Should we add a test case too ? I would suggest adding either a) network test using APIs b) lower_to_tir test to see if expected TIR is produced when using copy constant scheduler or c) use a TE-based test. For a), use simliar API like this to create a small graph : https://github.com/apache/tvm/blob/aa99699e532ec8c60afd552e951d238b481c4b5d/tests/python/contrib/test_ethosu/test_codegen.py#L162-L192 and then use https://github.com/apache/tvm/blob/aa99699e532ec8c60afd552e951d238b481c4b5d/tests/python/contrib/test_ethosu/test_codegen.py#L389 this for comparison. cc : @dchauhan-arm For b) you could use relay snippets such as : https://github.com/apache/tvm/blob/aa99699e532ec8c60afd552e951d238b481c4b5d/tests/python/contrib/test_ethosu/test_legalize.py#L62-L79 and then use a test like this : https://github.com/apache/tvm/blob/aa99699e532ec8c60afd552e951d238b481c4b5d/tests/python/contrib/test_ethosu/test_compiler.py#L25-L44 For c) https://github.com/apache/tvm/blob/aa99699e532ec8c60afd552e951d238b481c4b5d/tests/python/contrib/test_ethosu/test_scheduler.py#L155-L166 I think for the given bug c) is smallest reproducer if provide a diamond graph while a) being a more end to end test. cc : @lhutton1 @ekalda What do others think ? -- 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]
