That's usually how it's done in systems using int as well: internal calculations are done with as much precision as needed, and only the significant results are converted back to int using a defined rounding strategy.