Hzfengsy opened a new pull request #10158: URL: https://github.com/apache/tvm/pull/10158
`compact_dataflow` is a very strong constraint, which requires all the blocks in the sub-AST to be complete or reduction blocks. Currently, there are two primitives that use such property: - compute-at: the total scope is compact_dataflow. - parallel/bind/vectorize: the total scope is compact_dataflow. However, requiring all the scope to be `compact_dataflow` is too strict that some correct transformations are banned. (please see the added test cases). Here I updated the requirements: - compute-at: only requires the input block is complete or reduction - parallel/bind/vectorize: only requires compact_dataflow under the input loop. cc @spectrometerHBH @junrushao1994 @vinx13 -- 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]
