[clang] [llvm] [SROA] Vector promote some memsets (PR #133301)

[Shilei Tian via llvm-commits]

shiltian wrote:

> I assume AMDGPU does not support unaligned loads

We do support unaligned access IIUC, and it is actually required by HSA ABI. CC @arsenm 

https://github.com/llvm/llvm-project/pull/133301