[llvm] [AArch64] Add SVE shuffle optimization pass (PR #193951)

[Graham Hunter via llvm-commits]

huntergr-arm wrote:

> There was a suggestion to add a generic tbl intrinsic in https://discourse.llvm.org/t/should-there-be-an-architecture-agnostic-variable-vector-shuffle-operation/88936. It might be worth pursuing if it is useful for vectorization. In this case it looks like it is performing a `shift + and`? Some look like they might be more optimal with just the `and` or just the `shift`.

I've added a comment on that thread. I'll look into prototyping that intrinsic in a PR, since it looks like nobody on the thread has done so yet.



https://github.com/llvm/llvm-project/pull/193951