Thank you. A tiered JIT will indeed cause much more optimisations to be made possible.

I made a Qemu backend targeting LLVM before, but that turned out to be way too heavy to be usable, I wonder if it's worth revisiting that idea nowadays...