[llvm] [AArch64] Add @llvm.experimental.vector.match (PR #101974)

[Paul Walker via llvm-commits]

paulwalker-arm wrote:

I think we're on the same page and you may be correct in concluding the stock IR route will be cumbersome, but I think we should get to the point of knowing it to be true before picking a solution because otherwise we cannot be sure it's the correct solution.

The splats are trivial to match (especially if they're constant).  Likewise, an OR tree with matching roots doesn't seem that hard.  It's the predication that most worries me because that will likely mean instruction ordering matters, which can make matching awkward.

https://github.com/llvm/llvm-project/pull/101974