[Mlir-commits] [mlir] [mlir][scf] Add a scope op to the scf dialect (PR #89274)

[Gil Rapaport]

aniragil wrote:

Without understanding why this op is needed, I'm not sure why we're adding it right now. Can we get the end-to-end example? When and where will it be used?

[Sorry for the late response] Right, I uploaded this draft following [Renato's EuroLLVM'24 round table](https://discourse.llvm.org/t/notes-from-the-mlir-upstream-round-table-eurollvm-2024/78374/10), hopefully making the discussion on whether such an op is required a bit more concrete.
As @ftynse mentioned, one use case (and the original motivation) for this op is a proposed transform op that uses it in order to isolate transformations. The round table raised a second possible use case as a means to encapsulate graphs of ops (e.g. linalg.generic) in order to retain their idiom (in response to an issue raised by @mgehre).

https://github.com/llvm/llvm-project/pull/89274