[Mlir-commits] [mlir] [mlir][VectorToLLVM] add opt-in `enable-gep-inbounds-nuw` pass flag for `vector.load/store` (PR #202118)

[Federico Bruzzone]

FedericoBruzzone wrote:

> > Although the maintainers of the respective dialects will be there to offer their perspectives.
> > In my own small way, I'd like to help (study and) map the current landscape.
> 
> If you have the cycles, then please go ahead :)
> 
> You could post your findings either as a GitHub issue or a Discourse thread. Hopefully there won't be too many inconsistencies 😅

I'll try to do it as soon as possible :D

Silly question: is it correct for load and store to behave with different semantics depending on the dialect?
I'd rather open a discussion on the Discourse (and possibly an issue addressing that, if it's good practice).

The correct question to ask is:
"Is the current behavior of load/store in different dialects correct?"
or
"Do we want to align all load/store semantics?"

I'm asking you this as a novice.

https://github.com/llvm/llvm-project/pull/202118