[Mlir-commits] [mlir] [mlir][memref] Canonicalize memref.reinterpret_cast when offset/sizes/strides are constants. (PR #163505)

[Krzysztof Drewniak]

krzysz00 wrote:

I'd argue that the op was already invalid before canonicalization - just not verifiably invalid?

The size of a dimension - even a dynamic one - should always be nonnegative? 

So that c_-1 should be %runtone.var.with.sise or a `my.trully.arbitrary.size.marker`

https://github.com/llvm/llvm-project/pull/163505