@asmcall may be moved around/duplicated, since it's "only" a memory clobbering sideeffect. We can't mark inline assembly as volatile in LLVM-IR so preventing the moving may be a bit harder. Perhaps a pseudo-intrinsic that we noop afterwards is suitable? That may also help with #30 .
Ref https://github.com/maleadt/LLVM.jl/issues/320.
@asmcallmay be moved around/duplicated, since it's "only" a memory clobbering sideeffect. We can't mark inline assembly as volatile in LLVM-IR so preventing the moving may be a bit harder. Perhaps a pseudo-intrinsic that we noop afterwards is suitable? That may also help with #30 .