comaniac commented on pull request #10474: URL: https://github.com/apache/tvm/pull/10474#issuecomment-1058628299
Although I'm not 100% for sure, I'm afraid of performance impact. The tricky points are: 1. Although TOPHUB does out of date, many tuned schedules are still applicable. For example, the tile sizes tuned for Conv2Ds in ResNet on NVIDIA T4, V100, etc should still applicable, as AutoTVM and TVM CUDA codegne are rarely updated in the past year. 2. We do know that some ops (especially Conv2Ds on x86 and CUDA) do demonstrate obvious performance improvements with the TOPHUB schedules. General speaking, as long as we still support AutoTVM, we should not abandon TOPHUB. -- 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]
