If your inputs are single-precision float
, then you should be fine if you force double-precision arithmetic:
xSqr = double(x1 - x2) * (x1 - x2);
// ^^^^^^
If the inputs are already double-precision, and you don't have a larger floating-point type available, then you'll need to rearrange the Euclidean distance calculation to avoid overflow:
r = sqrt(x^2 + y^2 + z^2)
= abs(x) * sqrt(1 + (y/x)^2 + (z/x)^2)
where x
is the largest of the three coordinate distances.
In code, that might look something like:
double d[] = {abs(x1-x2), abs(y1-y2), abs(z1-z2)};
if (d[0] < d[1]) swap(d[0],d[1]);
if (d[0] < d[2]) swap(d[0],d[2]);
double distance = d[0] * sqrt(1.0 + d[1]/d[0] + d[2]/d[0]);
or alternatively, use hypot
, which uses similar techniques to avoid overflow:
double distance = hypot(hypot(x1-x2,y1-y2),z1-z2);
although this may not be available in pre-2011 C++ libraries.