So slower than tcc on O0 and slower than qbe on O1, with less backends, but tight llvm integration usable as baseline jit compiler, where it already outperforms Cranelift.