Once you've got an optimiser none of that really matters. Clang actually goes the other way, it emits IR with stack allocated variables (alloca), but LLVM's mem

Once you've got an optimiser none of that really matters. Clang actually goes the other way, it emits IR with stack allocated variables (alloca), but LLVM's mem2reg pass then attempts to convert those into single static assignment (SSA) form instead.