[llvm] [AArch64] Use SVE for load-extend-broadcast (PR #221689)

[David Green via llvm-commits]

davemgreen wrote:

> > Why "on V3AE" as opposed to all cpus with the available instructions?
> 
> In testing this was only found to be beneficial on certain CPU's. I've also renamed the PR to reflect this is applied to not just one CPU now.

It should be the other way around, if I understand this correctly. As there is a native instruction to support this pattern with SVE it should be the default codegen and CPUs for which it is slow for some reason should have it disabled. Using the existing target feature is probably the best bet.

https://github.com/llvm/llvm-project/pull/221689