[Mlir-commits] [mlir] [mlir][linalg] Constrain the parameters m, r in Winograd ops (PR #144657)

[Georgios Pinitas]

GeorgeARM wrote:

Is the implementation unable to functionally work with different `m`, `r` parameters?
If yes constraining is fine; but if it can then constraint semantics should be orthogonal. Otherwise we restrict experimentation and we might end up with a long list of configurations. Users using Winograd I presume need to be aware of what they are doing. If we want to ease we might have a utility function that returns some "safe", "good" parameters for a given config.
Thoughts?

https://github.com/llvm/llvm-project/pull/144657