This avoids going over a double precission float, which may loose precission
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>