[llvm] [AArch64] Use `AArch64ISD::UADDLP` for manual widening adjacent arithmetic (zext/shuffle combination) (PR #189255)

[Rajveer Singh Bharadwaj via llvm-commits]

Rajveer100 wrote:

> - add(shuffle_vec(zext(v0, ...), ...), shuffle_vec(zext(v0, ...), ...))
> - add(zext(shuffle_vec(v0, ...), ...), zext(shuffle_vec(v0, ...), ...))

In the two cases above, the second one we can detect the uzip1/uzip2 pattern but not for the first one.

Now that I think, we can convert the first one from `shuffle(zext)` to `zext(shuffle)` so that way we should always have the uzip form and then tablegen can pick up from here?

https://github.com/llvm/llvm-project/pull/189255