mbrookhart commented on issue #5647: URL: https://github.com/apache/incubator-tvm/issues/5647#issuecomment-632398831
Okay, I see the problem. I'm doing a post order traversal of the expression to find subgraphs to partition. That means it comes to the bias-add first, that matches the pattern, so it marks that subgraph as part of a fusion. Then it gets to the relu, ALSO matches it to the pattern, but can't fuse the nodes because they're already in another subgraph. I either need to allow growing fusion groups, which seems a little sketchy, or do a preorder traversal. I think pre-order traversal will be cleaner. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org