[llvm] [CodeGen] Simplify updateLiveIn in MachineSink (PR #79831)

[Jay Foad via llvm-commits]

jayfoad wrote:

> This patch avoids the problem I mentioned. However, it won't solve the problem, in my opinion. We don't want to have 0x00000000 lanemask value for general registers, don't we?

We're talking about special regunits that are only created to represent ad hoc aliasing between two registers. What should the lanemask be for such a regunit - 0 or -1 or something else? I don't think there is a single correct answer to that question. Maybe the lanemask should be poison and we should make sure that no-one ever depends on its value?

@qcolombet @MatzeB 

https://github.com/llvm/llvm-project/pull/79831