[llvm] [AArch64] Generate BSP instead of TBL for select shuffles. (PR #121474)

[Alexandros Lamprineas via llvm-commits]

https://github.com/labrinea approved this pull request.

The instruction sequences look better. Also BIF/BIT/BSL have higher throughput (4) than TBL (2), yet same execution latency (2).

https://github.com/llvm/llvm-project/pull/121474