[llvm] [PassBuilder] Move LoopFullUnrollPass to after vectorisation in the FullLTO pipeline (PR #196943)

[Momchil Velikov via llvm-commits]

momchil-velikov wrote:

> According to https://reviews.llvm.org/D102748, it may also make sense to do this for ThinLTO and non-LTO as well?

It may well make sense (I think it does, in principle).

However:
*  in that review there was an argument against
> The full unroll pass during function simplification is quite important for certain code patterns and serves a completely different purpose than the runtime unrolling that happens late in the pipeline. It is critical that full unrolling happens relatively early, so that scalar optimizations have a chance to work on the fully unrolled loop.
I don't think *this* patch is affected, since in the LTO pipeline between the old position of yje full unroll pass and the vectorizer there was only `LoopDistributePass`, no scalar optimisations.
 
* right now, I'm focused on the FullLTO pipeline and there's already a plenty of things to juggle with; but nothing in this work would prevent adjustments to the other pipelines in the future.

https://github.com/llvm/llvm-project/pull/196943