> calculating x * p/k from scratch whenever x changes doesn't always work either. You can still get significan rounding errors if the intermediate calculation of x * p is such a large number that it barely fits in a floating point.

Too late to edit that accidental markdown, so replying to myself