[llvm] [ARM] Speedups for CombineBaseUpdate. (PR #129725)

[Peter Smith via llvm-commits]

https://github.com/smithp35 commented:

For a non-expert this looks reasonable to me. I've made a couple of stylistic comments but nothing significant.

I guess the biggest decision point is whether 64 is the right choice. It seems to be the right choice for the problematic test case, and sounds plausible to be enough for unrolled loops.
I'm inclined to say let's give it a try and tune it there's any evidence of regressions.

https://github.com/llvm/llvm-project/pull/129725