spectrometerHBH commented on PR #14361: URL: https://github.com/apache/tvm/pull/14361#issuecomment-1480130861
> Yeah I agree. One issue that comes to my mind is that it might be critical to decide what subexprs to eliminate. If it's lightweight and inlinable to surrounding ops, then we should probably reject to eliminate the redundancy. In classical settings, the trade-off of CSE is to enlarge the live range of some vars, which might cause performance regression due to register spill. But for DL workloads, it's likely that the model simply can not be deployed given a certain amount of GPU Memory. -- 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]
