[llvm] Swap UnrollAndJam Pass to before the SLP Vectorizer Pass (PR #97029)

[Sjoerd Meijer via llvm-commits]

sjoerdmeijer wrote:

> Can't really comment on the change itself, as I'm not familiar with the pass.

Performance numbers shows that this doesn't regress existing benchmarks, and by moving the pass we enable more vectorisation. I think that makes this a very sensible change. 

@nikic 's suggestions are good things to fix. About compile-times and reuse of analysis results, I was curious if that is just a bit of trial-and-error, or is there a smarter way to see if a pass is recalculating its required analysis passes? Can this e.g. be dumped in debug logs? 

https://github.com/llvm/llvm-project/pull/97029