[Mlir-commits] [mlir] [MLIR][Bufferization] Fix IdentityLayoutMap allocation at function boundaries (PR #227253)

[Matthias Springer]

https://github.com/matthias-springer commented:

Can you share the before/after IR? I'm wondering if this is a bug in `castOrReallocMemRefValue`? That function is also called from `foldToBufferToTensorPair`, which is triggered from `to_buffer(to_tensor(x))` canonicalization. Is that canonicalization also broken?


https://github.com/llvm/llvm-project/pull/227253