I'm not sure I understand your question, but you want to evaluate the offsets outside the do-loop (i.e. at the time the instrument is called, not on every sample). A "sufficiently intelligent" optimizer could recognize that those entire expressions are not changing, and pull them out of the loop for you. Maybe someday...