[Mlir-commits] [mlir] [mlir][MemRef] Add support for emulating narrow floats (PR #148036)

[Andrzej Warzyński]

banach-space wrote:

> Since the memref types don't matter

Could you elaborate on _why_ that's the case? (*)For example, any kind of arithmetic on floats would be incorrect if we were interpreting them as ints - though I understand this patch only touches loads and stores.

I’m generally a bit concerned about the state of `NarrowTypeEmulation` in `Vector`. We should be cautious about how it's being extended. In my opinion, it currently lacks some of the rigor or design clarity we'd ideally want. Here's one case in point:

* https://github.com/iree-org/iree/issues/20645

I'm not opposed to this patch - the change itself makes sense. But if there are more changes coming in this area, it would be helpful to get a better sense of the broader direction. In particular, would you mind chiming in on this discussion?
* https://github.com/llvm/llvm-project/issues/131529

Thanks!

(*) Preferably in the summary :)

https://github.com/llvm/llvm-project/pull/148036