[llvm] [mlir] [OpenMP][MLIR][OMPIRBuilder] Add a small optional constant alloca raise function pass to finalize, utilised in convertTarget (PR #78818)

[via llvm-commits]

agozillon wrote:

> > Just a small ping to check if you're happy with the fix as is @kiranchandramohan or if you'd prefer me to dig into the block merge/splice method to see if it's feasible?
> 
> No need to pursue this.
> 

Sounds good, if it ever becomes necessary to pursue the other option I am more than happy to do so.

> I have couple of questions inline. Also, would this be better as an LLVM pass?

I would like to avoid it being an LLVM pass for now if that's at all possible, it's a rather small/trivial piece of functionality that only gets performed on the Target/kernel entry function at the moment, I'd rather other people not start (or at least be able to) running it outside of the intended use-case for the moment. I think the easier to revert it is the better it is for the time being if we have to look at other options. However, if it expands to more complex use-cases (e.g. dynamic allocations or non-foldable constants, although, if we have to cover these cases I'd lean towards the entry block splicing or an alternative) then certainly it should be moved to an LLVM pass.

https://github.com/llvm/llvm-project/pull/78818