[Mlir-commits] [mlir] [mlir][tosa] Fold PadOp to tensor operations (PR #132700)

[Georgios Pinitas]

GeorgeARM wrote:

Ouch, you are right.
I actually think that this a rather problematic behaviour; it can limit folding in many cases.
Even an optional pass is still problematic.

**Options:**
- Change the canonicalisation to work under the condition that it doesn't break the expectations of the operator
- Put it on a separate pass with an option to allow breaking the limitation if requested
- Lift the limitation from the spec

https://github.com/llvm/llvm-project/pull/132700