================ ---------------- Keenuts wrote: I think we should a target-dependent alignment. Having an explicit alignment would not be lower-able to logical SPIR-V for alloca. I'll add a note in the docs https://github.com/llvm/llvm-project/pull/186811