[llvm] [AArch64] Don't run loop-idiom-vectorize pass in the O0 pipeline (PR #156802)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> I may be missing the broader context, while I do see the tagged issue. Is this purely because x86 does not run the pass? How do we decide what goes in O0?

As little was possible. The passes that are run at -O0 are maintaining historical semantic accidents, like always-inliner and should include no optimizations 


I'd expect the pass builder to not call this hook in the first place, rather than requiring all targets to check this 

https://github.com/llvm/llvm-project/pull/156802