[llvm] [AArch64] Skip storing of stack arguments when lowering tail calls (PR #126735)

[Guy David via llvm-commits]

guy-david wrote:

Wasn't able to implement a full solution for GlobalISel and the lowering still emits a load-store for `i1` passthrough arguments. The problem is that two truncations are introduced in CallLowering which make it difficult to analyze whether the the value that is being stored is the same one that was loaded.

I thought of pushing through with KnownBits, but unfortunately an `i1` and an `i8` are indistinguishable at this stage and the expected constant is different in each case. I'll open an issue describing it.

https://github.com/llvm/llvm-project/pull/126735