[llvm] [AMDGPU] PromoteAlloca: flatten homogeneous structs to vectors (PR #217055)

[Domenic Nutile via llvm-commits]

saxlungs wrote:

Actually, here's an end-to-end HIP godbolt that more explicitly shows the issue I'm trying to solve with this: https://godbolt.org/z/7vYb1G917

https://github.com/llvm/llvm-project/pull/217055