f2013519 opened a new pull request, #14832: URL: https://github.com/apache/tvm/pull/14832
Currently during the ConvertLayout pass, whenever there is a nn.batch_flatten op, additional layout transforms are inserted before the nn.batch_flatten op. This may not be efficient. In addition, for BYOC targets (which expects ops to have a certain layout), insertion of these transforms can create additional overhead - especially if they cannot be handled by target which then creates additional compiler regions. This can be avoided by propagating the updated layouts to the nn.batch_flatten op. -- 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]
