Can you elaborate on the extent to which you use LLVM optimizations? Is it comparable to e.g. clang's -O1/2/3?