[llvm] [AArch64] Replace expensive move from wzr by two moves via floating point immediate (PR #146538)

[Julian Nagele via llvm-commits]

juliannagele wrote:

Thanks! And you were right about not needing new patterns, so I've removed the ones I added and put the existing `INS .. WZR` ones behind a feature and added that to the cpus you mentioned.

> > Sure, do you maybe have an example I can work off, not sure how to go about this.
> 
> I think that DisableFastIncVL/HasFastIncVL shows how to add a subtarget feature and make patterns dependant on it. The old patterns can be made dependant on it, I believe, and it can be added to the processors that would not benefit. (FastIncVL is perhaps a bit odd as it doesn't appear to be used anywhere).



https://github.com/llvm/llvm-project/pull/146538