[llvm] [AArch64][CodeGen] Don't try to compute stack addresses in xzr (PR #213009)

[Eli Friedman via llvm-commits]

https://github.com/efriedma-quic commented:

Are we actually generating correct code for flag-setting operations involving frame indexes, with this patch?  There is no flag-setting variant of addvl, so I think we're getting condition flags from the wrong instruction.

I think I'd be happier if we were even more conservative here, and just completely bail out of the transform when frame indexes are involved.

https://github.com/llvm/llvm-project/pull/213009