following up on this, I think we are in broad stroke agreement that we can achieve our goals with blocl/fn attributes in IR as well as builtin assume. As a result, my original blocker for the RFC has been resolved, would still be great to work together to flesh out the details of schedule primitives and how do they interact with the rest of TIR scheduling, but I somewhat think they can be done separately and we don;t need to nail down the details of primitives.
The schedule primitives can be done relatively independently as long as we agree on the principle that: - Transformations do not change the function interface behavior - We decouple the graph level decisions into two steps: local decision + rewrite We can explore possible options as long as the IR spec remains stable, if there is a need to update IR itself or meaning of attribute, we can come back and discuss again -- Reply to this email directly or view it on GitHub: https://github.com/apache/tvm-rfcs/pull/77#issuecomment-1180606480 You are receiving this because you are subscribed to this thread. Message ID: <apache/tvm-rfcs/pull/77/c1180606...@github.com>
