[Mlir-commits] [mlir] [MLIR] Improve in-place folding to iterate until fixed-point (PR #160615)

[Matthias Springer]

https://github.com/matthias-springer approved this pull request.

This change makes sense to me.

The same strategy could potentially be used in the greedy pattern rewrite driver to improve efficiency: we could try to fold the op in a loop until no more folding is possible. This may be more efficient than pushing/popping the op onto the worklist. (We still have to put the op on the worklist after the last successful folding, so it's unclear if this actually improves compilation time.)


https://github.com/llvm/llvm-project/pull/160615